Archived Theses

This master’s thesis explores methods for mouse input in the Augmented Reality (AR) environment. A prototype is developed and implemented, allowing menus to be controlled for interacting with smart home devices using hand gestures and a Magic Leap2 headset. The main goal is to develop a userfriendly interaction technique that enables users to control various menu elements through simple hand gestures. To achieve this, techniques for visualizing and tracking hand gestures, hand-based raycasting, combining hand and eye movements, and determining the position of the wrist in relation to the controlled menu element are integrated. The prototype was tested in a controlled environment, where most requirements were met, but potential for improvement was also identified, particularly in real-time visualization and user-friendliness under different lighting conditions. This work contributes to the exploration of new interaction possibilities with menus for controlling smart household devices and provides a foundation for future developments.

Pest counting is crucial in agriculture and science to assess the condition of plants. This counting process is very time and labor-intensive. This study examines how augmented reality glasses can be utilized for this application and compares them with a smartphone application and the classic manual counting method. The results show that there are no significant time differences in the execution time between the techniques, but that the role of humans is fundamentally changed by the introduction of new technologies. In particular, the AR glasses show promising performance compared to the smartphone, although the glasses were unable to capture evaluable images. The error rate when interacting with the glasses is low, but further research is needed to make reliable statements about this. Additionally, the application of object recognition AI for automatic pest identification is being investigated, and it is concluded that the model developed is unsuitable for this experiment. This study was conducted with participants from the plant science, potentially providing a different perspective on the technology. Future work should take these limitations into account to further improve the efficiency of pest counting in agriculture.

As part of this thesis, a prototype is being developed that makes it possible to control a smart home using a Head-mounted Display (HMD) in augmented reality. For this purpose, the devices in a typical smart home were divided into categories. This categorization was validated in a survey of 28 participants. Attention was paid to typical distance to the devices, interaction time and frequency of use. Selection techniques were then developed for these categories, which were implemented in a software prototype in the next step. An additional input modality is added to the HMD used. This is realized via a touch sensor on the frame of the HMD and transmitted to the HMD via Bluetooth Low Energy using a microcontroller. At the end, a comparison is made with an existing prototype from a previous thesis.

A study examines the performance of the participants in terms of selecting objects in Augmented Reality (AR) with the input methods eye-tracking, headtracking and hand-tracking. This study concludes that eye-tracking is best suited for accuracy. However, head tracking is the most popular. Hand tracking is less suitable due to poor accuracy and reliability. Based on this study, a prototype was developed that allows smart home devices to be controlled using AR glasses. This prototype uses head-tracking as an input method and is based on two modes. The first mode allows users to look around and the second mode is used to select objects. The second mode is activated by blinking twice. The prototype was then evaluated in a second study and compared to an established method of controlling smart home devices. This study comes to the conclusion that the prototype has the potential to replace the established method, but in its current state, is significantly worse than the established application. However, most aspects of the user interface received positive feedback. Only the use of blinks to switch to the second mode received negative feedback. This also contributed to most of the prototype’s poor performance.

With the rise of technology, small input devices can compute more advanced applications. This makes smartwatches more desirable than ever. Smartwatches imitate the shape of classic watches and use their small touch screen for input. That turns into a problem regarding screen occlusion. This work attempts to overcome the screen occlusion problem by extending the touch input from the screen to the smartwatch case. A prototype with capacitive sensing allows touch input on the side and on the band of the watch. By utilizing a self developed software, recognizing and categorizing gestures performed on the prototype is possible. This system was tested and compared to touch in two experiments with 12 participants. In a first experiment the prototype was applied on a map navigation task. Two input techniques were implemented using the system. The first technique uses only a vertical sensor and the second technique uses a vertical and a horizontal sensor. Results show touch was significantly faster then both input techniques. Touch was 25.56 % faster then the input technique including two sensors and 40 % faster then the input method with one. Nevertheless both avoid screen occlusion and the technique utilizing both sensors was rated the highest by the participants. Comparing regular touch input between different screen sizes showed significant sensitivity towards the size of the screen. Touch, on a display size of 35.56mm, performed 40% faster then on a 30.48mm display. The second experiment was designed as a list scrolling task. Only a vertical sensor was used from the prototype. The results of this study also showed significantly faster task completion times with touch. For this task touch was preferred. An exception was ascertained when the target distance of an entry was 40 or 80 entries away, within unsorted lists. Users sometimes missed the entries using touch by blocking the visual feedback. This indicates that screen occlusion is a problem. With that in mind, the proposed input method shows potential to improve smartwatch interaction with small touch screens whenever screen occlusion is a factor or when the input space for touch is insufficient.

Augmented Reality (AR) is a technology that overlays digital information, such as images, videos, or 3D models, onto the physical environment in real-time. Developing user-friendly interactions with AR content is challenging and necessitates a well-balanced design of gestures, speech, and touch. The present thesis focuses on the „Design of Menus for Online Researches Using Handwritten Information in Augmented Reality“. The primary objective is to facilitate whiteboard usage through AR-supported Google searches of handwritten text and drawings. A user-friendly control mechanism is developed to ensure efficient navigation within search results. An in-depth analysis of existing technologies and methods for integrating Augmented Reality, handwritten recognition, and online research is conducted. Based on these insights, diverse design concepts are formulated to enhance interaction between users and digital information in the AR environment, emphasizing user-friendly controls. These concepts are implemented in a prototype and subsequently evaluated in a study for performance and user-friendliness.

This work dealt with the development of a haptic bicycle navigation system. The goal was to create a complete navigation system that does not rely on the use of audiovisual feedback. Haptic feedback was enabled through electrotactile signals. A wristband prototype based on previous work was used for providing the feedback, which consists of nine individual electrodes capable of delivering electrotactile signals to the skin. The layout of this prototype was used to design the navigation maneuvers. These were generated using the Mapbox SDK. This SDK allowed for the development of a complete navigation app. The route was based on individual waypoints, serving as the foundation for the individual maneuvers. The navigation did not rely on traditional maneuvers but rather aimed to represent the angle between the current position and the next waypoint. To test the functionality of the system, an evaluation study was conducted. The study assessed the usability of the prototype and examined the influences of the environment on the usability of the system. Four male and four female participants were involved in the study, ensuring a diverse participant group with an equal gender and age distribution. It was evident that the system provided accurate navigation, although its accuracy decreased with increasing speed. This work can serve as a foundation for the development of novel navigation systems based on electrotactile feedback.

Wearable digital devices are commonly used nowadays. While wearables are typically used to support the wearer in a variety of tasks, smartwatches and other such devices also offer different opportunities. In addition to content meant for the wearer of the device, the wearable can also be used to display information to different people glancing at its display. In this work, a digital brooche for use as a public display is designed and evaluated. For this purpose an online survey is employed to collect data on social acceptability and usefullness of such a device. The data gained from this survey is then used to create a prototype. The visibility and noticeability of information displayed on the prototype is then evaluated in a study. Finally the results of this work are displayed and discussed, pointing out and elaborating on different tasks and difficulties in the design of public displays.

Augmented Reality (AR) is a technology which integrates digital information into the real world using digital devices. One of the challenges is to create a natural and intuitive interaction between the user and the system. To achieve this, different input methods can be used, such as voice, gesture or touch. The present thesis focuses on using handwritten commands, as input method for controlling AR-applications. The aim is to evaluate whether there is a need for handwritten commands for interaction between the user and the AR system. Based on this, possible designs will be explored which contribute to make the technology as intuitive as possible. The use case considered here, examined the use of AR glasses for solo whiteboard work, as a first idea. The thesis includes a preliminary study to evaluate the use of handwritten commands as well as the development of the possible design and implementation of such handwritten commands. Based on the preliminary study, the implementation for a specific functionality was then carried out as a minimal example (prototype). This prototype was then tested in a subsequent study with regard to usability and natural interaction of the handwritten commands with the AR application

Smartwatches and similar wearables are gaining popularity and are in direct contact with the skin. The focus of this work is to support visual feedback on a smartwatch by integrating electro-tactile feedback. For this purpose, a prototype is developed consisting of an electrode array and a circuit to activate individual electrodes. This prototype can be mounted on the back of a smartwatch and is in direct contact with the skin. To investigate the functionality of this prototype in combining visual and electro-tactile feedback, a study was conducted with 15 (8 male 7 female) participants. The difference between the genders played a significant role in the calibration of the prototype. In one part of the study, participants were asked to view GIFs on the smartwatch while simultaneously feeling the corresponding electrotactile feedback. It was found that GIFs with powerful and fast movements of the electrotactile feedback were perceived better and evaluated more positively than GIFs with slow movements. Overall, 98.3% of the participants perceived the electrotactile feedback positively and felt it well. These results demonstrate the potential of electrotactile feedback in combination with visual feedback to improve the user experience of wearables.

This thesis focuses on the application of deep reinforcement learning in conjunction with electrical muscle stimulation (EMS) to achieve precise control of the human hand. The primary goal is to develop an automated method for calibrating EMS devices to perform hand movements using learning-based approaches. To achieve this, several research questions were posed regarding the characteristics of the controller to be found. Systematically answering these questions led to the development of several artifacts that enable the iterative construction of better models. First, classical, non-trainable models were developed to test the assumptions made in the initial phase of this work. Neural models that control movements by turning a single EMS action on and off were then explored. Subsequently, models that can use the entire EMS parameter space (intensity x pulse width) were also explored. To this end, a procedure was introduced to create personalized EMS action spaces that can be used to map network actions in regression settings. Different reward functions (homographic, asymmetric, stepped, and Gaussian) were developed and tested to determine their impact on the resulting policies and the training process. Results were then compared to the human baseline using data collected with 6 participants. The work also sheds light on the limitations of the model and possible areas for further research.

In this paper, concepts for designing interaction methods using Bluetooth’s Direction Finding methods, Angle of Arrival (AoA) and Angle of Departure (AoD) are presented. In particular, the focus is on studying the capacities and limitations of Bluetooth-AoA. Based on the results of these studies, arm gestures were designed to be used to interact with Bluetooth devices via the measured angles of arrival. In a user study with 16 participants, these gestures were evaluated in the context of targeting accuracy and measurement accuracy for the angles of arrival. For the experimental studies, prototypes were developed to enable the exchange of direction finding packets, allowing the use of Bluetooth AoA functionalities. Results have shown that Bluetooth device selection using a pointing gesture is relatively reliable with a detection rate of at least 70%. For the performance of the pointing gesture, it has additionally been found that the targeting accuracy of users is less accurate with the left arm. In contrast, the measurements of the angle of arrival are less accurate on the right arm. Despite a deviation of about 3° from the actual angles, the gestures using Bluetooth AoA offer robust possibilities for interaction with Bluetooth devices.

In this thesis a smartwatch prototype is created which automatically displays supporting information during a conversation. For this purpose, an existing hardware prototype with two additional screens aligned for the participants of the conversation will be extended by a speech recognition system. The speech recognition continuously evaluates the content of the conversation. If something is recognized that can be displayed, it is displayed including relevant contextual information on the additional screens of the smartwatch, visible to all participants in the conversation.

This work is a continuative work. In the previous work, haptic mouse feedback using electrical muscle stimulation (EMS) on the forearm was investigated. The stimulation of the muscles should restrict the movement of the hand in the direction of the thumb as well as the little finger. In this work, the muscles are stimulated, also with EMS, on the upper arm to affect the movement of the hand from the body or to the body. In a study, the muscles are evaluated for their interpretation, recognition of different values, and scalability on a slider on a graphical user interface. At the end of the study, a combination of both works is tested using a selected application scenario. Finally, the resulting data is evaluated.

Counteracting the formation of filter bubbles and their role in the spread of misinformation remains a hot topic in the context of online platforms. Several approaches towards achieving a more balanced and sensible interaction with content on (social) media services have been proposed in the past. Most rely on self-control and a willingness to confront oneself with dissenting views. There remains a knowledge gap concerning whether the relevant target audience would be willing to employ such interventions. Additionally, a real risk is present of achieving the opposite effect if such interventions are deployed, as confrontation with dissenting opinions can serve to harden one's views. Similar questions can be found in the area of sustainable HCI, in which a large set of interventions towards a more sustainable lifestyle have been proposed. Here, the question of validating the effectiveness of research in this area is an ongoing issue as well. This thesis concerns with the question to which extent such interventions can serve a purpose. That is, whether the relevant audience would employ them and whether they aid in shaping a balanced worldview. For this, an extensive literature review needs to be conducted to collect large selection of proposed interventions. Its likely that one has to segue slightly into the basics of sociology and psychology as well. Depending on the findings, either a survey or a small scale study can be conducted for the purpose of verification or contextualization.

After electrical muscle stimulation (EMS) has been used successfully in medicine for years, more and more applications in other areas have recently been researched. In human-computer interaction, EMS can help with the output of complex information without the user having to pay attention to auditory or visual feedback. However, the increasing complexity of the data also poses increasing challenges for the controller. In this work, several existing strategies in wrist control are extended. For this purpose, approaches that use reinforcement learning are analyzed in short experiments. At the same time, approaches without neural networks are also optimized so that a comparison with regard to the quality and the training duration of the controls is possible. It is shown that neural networks can represent an option for the realization of the control, but are not always the optimal solution. Instead, simpler controllers tailored to a specific problem are often equivalent or even better. These then also work without prior training. Binary controller deviations were less than one degree for constant functions and up to eight degrees for high-frequency sinusoids. A high time dependency was determined in reinforcement learning, whereby the deviations were more than ten degrees even for constant functions.

This work compares the concept of a vertical computer mouse with the classic computer mouse in terms of the user’s muscle activity. For this purpose the computer mouse models Logitech MX Master 3 in the classic design form and Logitech MX Vertical in a vertical design form are compared under various criteria within a user study. The first task requires drag-and-drop actions over long distances and the second task requires computer mouse movements over short distances but with many changes of direction. During the execution the screen contents, mouse hand of the participants, muscle activity of the flexor carpi ulnaris muscle, computer mouse interactions, mouse typing interactions and results of the test executions are recorded and evaluated. The computer mouse models are compared in the disciplines speed of task processing, muscle activity of the flexor carpi ulnaris, cursor speed, subjective evaluation of the participants using a questionnaire based on the System Usability Scale, and precision. The Logitech MX Vertical was rated with a score of 60.17 and the Logitech MX Master 3 was assessed with a score of 67.83. The user study showed that the central tendencies of the muscle activity measured in the study conducted must be considered the same for both computer mouse models.

For an ever increasing amount of people, a computer is a device they use on a day-to-day basis. In order to use a computer, an input device is needed, which is often a mouse. However, it lacks a form of haptic feedback, which can transfer information from the graphical user interface to the user. Therefore, in this thesis a prototype will be designed which enables haptic mouse feedback via electrical muscle stimulation. This will alter the position of the users wrist and the user perceives Force Feedback. The developed prototype is integrated in a application scenario, which includes the placement of an object on a graphical user interface. The prototype is tested in a user study, which focusses on every-day usage, in particular the precision of the placement and compares two stimulation modes. One stimulation mode is significantly more accurate. Afterwards, the prototype will be evaluated with the data from the study.

In this work, another input modality for a smartwatch interaction is implemented and evaluated. The input modality is a gesture recognition system that uses electromyography (EMG) to measure muscle activity in the forearm and can recognize hand gestures. The advantage is that only one arm needs to be used as opposed to touch input, since the EMG wristband and smartwatch are attached to the same arm. The special feature of the EMG prototype used in this work is that it sequentially triggers the individual electrodes and then measures the activity via an external EMG sensor. The user study conducted for this work with 15 test subjects showed that the gesture recognition could recognize the gestures with 83.2% accuracy and the reaction time of the test subjects for gesture recognition (5820ms) is slower than that of touch screen input (4661ms). This is due to the recognition time of 1080ms for a gesture. Subtracting this for gesture recognition, there is no significant difference between both input modalities. Since gesture recognition results in fewer errors during the execution of other tasks, it can be assumed that the one-armed interaction offers an advantage over touchscreen input.

This work investigates the noticeability of different output modalities on the earlobe and meaningful applications in everyday life. Noticeability is composed of two aspects. The error rate describes the frequency at which active stimuli were not perceived. On the other hand, noticeability consists of the reaction time with which a stimulus is perceived. For this purpose, prototypes were developed that can be attached to the earlobe. These prototypes include a total of 9 modalities: Electrotactile feedback, Vibration, Thermal Cold, Thermal Warm, Sound, and Poke form the private modalities that the user perceives himself.Display, Light, and Sound Public are public modalities used to convey information to others around the user. A survey was conducted to gather initial impressions of each modality and to ask about specific usage scenarios. The results could later be compared with the surveys from the laboratory study. The laboratory study showed that there were significant differences in terms of the response time. The thermal cold and warm feedback showed no significance among themselves, but were significantly slower than the other private modalities. Furthermore, vibration feedback was significantly faster than poke feedback. For the public modalities, there was significance regarding response time between the display and both other modalities. The display was significantly slower. There was no significance between Light and Sound public. The error rate was 0 % for most modalities. The thermal cold feedback had an error rate of 27.78 %, the thermal warm feedback had an error rate of 3.89 %, the sound had 0.56 % and the display had 22.22 %.

Headphones allow defining your own soundscape and privately listening to audio sources in public. Active noise-cancelling headphones can even reduce external sounds that intrude into this soundscape. However, to others, the state of a person wearing headphones is unclear (is s/he listening to music? may s/he be spoken to?), which is socially problematic. The goal of this thesis is to explore whether displays integrated in headphones may help to communicate the state of the user to other people. What should be depicted on the display to indicate the state? What about privacy concerns? How should the headphones sense the state of the user? How could the user explicitly provide the content to be shown? This thesis comprises brainstorming about scenarios, sketching and designing potential visualizations on earpieces, asking potential users about the prospect of such display-enhanced headphones, and (if done as a Master's thesis) implementing a prototype.

The fact that the finger covers the screen during touch input and the user can no longer see the screen content is still a problem when using smartwatches. Many smartwatches have alternative input possibilities to the touchscreen to prevent this problem. These include the rotation of the crown or bezel to interact without covering the screen with the users finger. This thesis will analyze already implemented interaction techniques and extend them by using a trackball as input device. Therefore, a trackball crown should be created for an Android smartwatch. Focus of this thesis is also the evaluation of a trackball as input device for smartwatches and the analysis of the user acceptance in user studies.

This thesis is concerned with the development and evaluation of methods for assessing the accessibility of content developed for the web. While technical standards and guidelines towards enabling the development of accessible web content have existed for some time, research into tools that provide an insight into the fulfillment of said guidelines is still ongoing. Two examples for such tools are WAVE and TAW. Yet these tools can only give informal feedback and therefore may lead to wrong conclusions about how issues should be remedied. Additionally, in recent years additional challenges have started to be discussed, such as the problem of seizure inducing animations, which extend the problem space. In a novel approach the limitations affecting accessibility are to be made tangible for able-bodied people in order for them to gain deeper insight into perception of their content under different constraints. This is to be done through what is essentially a simulation of the perception of web content under different disabilites. By developing the tool as a browser extension, regular interaction with web content can still be provided. The theory hereby is that, through experiencing possible problems firsthand, developers can immediately spot and prioritize and fix accessibility issues themselves. Through a user study the developed system is to be evaluated regarding its effectiveness when compared to informal approaches such as WAVE or TAW.

In previous work we’ve evaluated methods to qualify and cluster Twitter profiles using the platform’s social graph as a basis. While yielding favorable initial results, this technique is only able to perform a very broad classification. This is attributed to the fact that several other factors need to be taken into account when categorizing online profiles. Furthermore, the approach requires expensive pre-processing and is therefore unsuitable in applications that involve live data. In this work you will be given an extensive database of Twitter profiles and tweets, with the aim to incorporate a more extensive chunk of the data into the clustering approach. Through the application of machine learning, the required time of classifying unknown profiles can be significantly reduced once a model is trained for the task. As this approach has the potential to enable usage of information in real-time, a separate demonstrator application is to be developed to make use of this ability. The application’s aim is to give a live overview of Twitter postings in regards to their corresponding communities. For this, Twitter’s filtered or sampled stream APIs may be used in combination with the developed classifier to provide live data. Through an auxiliary user study the effect of this visualization method on the perception of content can be compared to that of a regular time-sorted list.

Publications in HCI's subfield Sustainable HCI strive to present systems with a positive ecological impact. Common themes have included persuasive technology and ambient awareness as means to foster changes in personal behavior. However, as effects are yet to be widespread enough to positively influence global emissions and resource usage, other works criticized such approaches and ascribe a different role to HCI, in which more focus should be placed on the supply aspect of devices. Recent works also called for an application of the principles behind permaculture, in which waste is essentially avoided in favor of a circular resource usage. A more extreme lens is provided by Bill Tomlinson et al. with the notion of Collapse Informatics, in which a deliberately negative speculative future is used as a framing device for sustainability efforts of HCI and ICT in general. By imposing an assumption of extreme limits on resource availability, taking the complete supply chain into account automatically becomes a necessity when assessing designs of digital systems. In this thesis, the aim is to provide a comprehensive framework to help evaluate all facets of sustainability focused projects, in which their complete supply chain and lifecycle is taken into account. With this framing, a critical look at a representative subset of existing literature in the area of SHCI is to be provided. Furthermore, the framing is to be used as a tool to evaluate the viability of 2 to 3 existing and influential HCI/ICT projects. Their shortcomings are to be identified and speculative designs that help fit these projects into the developed framework are to be given. This, in turn, serves to identify user-visible aspects of such developments and is to provide a baseline for future work in which user concerns of such systems are addressed.

The role social platforms play when it comes to radicalization on the internet is a hot topic. Increasingly segregated sub-communities ("filter bubbles") serve to reinforce fringe opinions and don't allow for healthy exposure to sentiments expressed elsewhere. Existing work has already been done to identify such bubbles within the social graph of Twitter. The goal of this thesis is to design an interface that enables an interactive exploration of this graph, with a focus on social bubbles and their interconnections. This includes researching methods to map and present the data and to consider which additional information could be displayed. Furthermore, a user study will have to be conducted to evaluate the design's expressiveness and intelligibility.

The content provided to users of social platforms is pre-filtered based on past behaviour on the platform. It is subject to selection biases of individual users, thus preventing any confrontation with divergent opinions. This thesis concerns with the design of an interface that makes users aware of overly selective exposure in their social feeds. The aim of this interface is to subtly steer users towards a broader exposure via nudging. Said design nudge should incentivize users to counteract increasing disconnection of sub-communities on social platforms. Furthermore, a user study is to provide insight into the acceptance of such a feature and the agreement with its feedback.

Searching for new control and input methods for small computers, such as wearables, one possible option is gesture-based control. This requires capturing hand gestures by a computer. In this work, an electromyography-based approach is taken to detect hand gestures using a provided EMG wristband prototype with eight electrode units. Software for the Arduino-based prototype and software for hand gesture recognition are developed. The modular recognition software processes the incoming signal from the prototype, recognizes the underlying hand gesture, creates recordings with hand gesture data sets, trains various classification methods, and evaluates these methods. In nine experiments with a total of two participants, the different classification methods Logistic Regression, Support Vector Machine, Naive Bayes, Random Forest Classifier, and Multilayer Perceptron Classifier are investigated. The hand gesture recognition is improved by means of the experiments by restructuring the input data and optimizing the parameters of the Random Forest Classifier resulting in an accuracy of up to 93 % for five different hand gestures using the provided prototype

The output capabilities of smartwatches are limited by the design and the used technologies, e.g. vibration and displays. We envision the use of electrotactile feedback to expand the output space of smartwatches. A working prototype that can present various sensations via electrotactile feedback is available and should be used to design and implement mobile scenarios that benefit from the enhanced output capabilities. Scenarios could be in the context of notifications, navigation, mobile gaming or etc. The thesis should start with a focus on literature to discover potential application scenarios. Then the thesis should create concepts for the discovered scenarios and after that at least one scenario should be implemented and evaluated with a few participants.

Smartwatches are becoming more and more popular. But so far the screen size is limited to the watchface. In this work, a smartwatch watch strap has to be developed to extend the smartwatch display from the watchface to the watch strap. For this purpose, flexible displays will be used which can display additional information or notifications. The analysis of the design space on the watch strap, as well as the possible areas of application are to be identified and implemented. In addition, haptic feedback along the watch strap should be provided to inform the user about specific information (and their position) on the watch strap display.

This thesis will deal with ear clips as output devices for notifications. Therefore an ear clip prototype will be created and evaluated with different feedback modalities (e.g. vibration, sound, LED, ...). The focus is on the evaluation of these modalities and their comparison. The ear clip should be able to connect to any Android device via Bluetooth and inform the user about notifications received on the Android device. Knowledge in Android programming is recommended.

Smartwatches and fitness trackers are getting more and more popular in public. With the skin contact provided by these wrist worn wearables, the usage of electrotactile stimulation in notification context becomes more simple and reasonable. This thesis investigates what sensations can be evoked by electrotactile feedback on the wrist skin and if they are suited to provide notifications. Therefore a modular, round, watch-like prototype was constructed to easily evaluate different electrode designs. The final version contains two opposing rubber electrodes on the bottom of the prototype. To apply electrotactile stimulation to users a RehaMove$3$ device, a medical certified configurable electrical stimulator, was used. It holds up a vast parameter space compared to other off-the-shelf devices. The RehaMove 3 does not offer wireless communication out of the box. For the mobile control of the RehaMove 3, a microcontroller was connected, a ported control library was finalized and an Android application was written. A study with 13 participants was conducted with the mobile prototype to evaluate the combinations.

In this project a pest monitoring system will be implemented using a Raspberry Pi, some LEDs and an IDS camera. The monitoring device is used to attract insects using light frequency of the LEDs and to monitor them using a camera. The insects will be detected using the You Only Look Once (YOLO) algorithm and further classified using Support Vector Machine(SVM). Two insect species (Aphids and Thrips) are identified with the method mentioned and the classification is done into three classes (Class 1: Aphids, Class 2: Thrips, Class 3: Others). The system will provide feedback to the user via an Android application. The real-time insect population will be recorded and displayed in a graphical form in the Android application. Furthermore, in the Android application, the user is able to set the type and frequency of the notification, either notifications in regular interval (daily or weekly basis) or warnings when a certain threshold is reached. The threshold mentioned is a population value that is either predetermined via experimental executions (eg. 80% of the maximum population) or inserted the threshold manually from the user. The setup feasibility of the monitoring system and the procedural execution in Android application are investigated in the user study. A simulated garden environment is conducted to ensure the users are able to use the system in a correct manner. Before the execution of the user study, a user manual for the monitoring system will be written for the user understanding. In the end of this thesis, an analysis/ a data tabulation that consists of usability and user-friendliness will be included from the survey conducted.

An Android application will be developed for amount-counting task of the pests in green house for this project. The whole system consists of two parts - mobile device with application and the server. User can use the application to take photos for the leaves of the vegetables and then these photos will be sent to the server for image recognition and object detection so that the number of pests will be automatically counted. By the server side, an image recognition algorithm will be run to calculate the number of the pests (and also with classifications of the pest types), after calculating server will send the results back to the user side and user can easily get the total quantities of different pests. Based on the result numbers user can take actions if the value of the pests’ quantity is beyond the threshold to protect the crops. Furthermore, we can see the data in server via ssh in the computer to analyze the data for experiment or somewhat. At the end of the project, a user study will be conducted, before the user study one user manual is required to be written and then user can read the manual to understand how to use the application.

Electrical Muscle Stimulation (EMS) is being researched in many different applications in the field of human-computer interaction, such as learning to play an instrument. Current problems refer to a only rough control of the muscles. Therefore, a solution strategy using neural networks is examined here. The research question is: How well are neural networks suitable for controlling an EMS application? To answer this question, the basic problems of EMS are first analyzed. Thereby an approach is presented which is based on the combination of Reinforcement Learning with neural networks. Experiments show that Recurrent Neural Networks (RNNs) are superior to feedforward networks. Analyses show significant differences in controlling of different angles. Therefore, comparisons between controllers for individual angles must be made. A comparison of PID controller and RNN shows an average increase in precision of 4.17°, in favor of the RNN. A comparison between PID controller and Bidirectional Recurrent Neural Network (BiRNN) shows an average increase of 7.63°. A comparison between RNN and BiRNN shows an increase of 3.46° by the BiRNN.

Nowadays, most people own at least one smartphone and at the same time smartwatches and other wearables are becoming more and more popular. These devices send a variety of notifications to the wearer, usually in the form of vibration, sound or a status LED. Further, many users prefer invisible feedback, so that only the attention of the addressee and no third party is attracted. Accordingly, vibration feedback is used to a large extent, but this feedback is not sufficiently distinguishable. In this thesis, a wristwatch for mobile use is designed to evaluate an alternative invisible notification method using electrotactile feedback. By adjusting the parameters of the electro-tactile feedback the creation of different sensations is investigated. Potentially distinguishable natural notifications such as vibrating, itching or prodding are generated. A wristwatch prototype with corrosion-resistant gold-plated electrodes was designed, which were produced in a new electrode manufacturing process. Furthermore, a calibration procedure is presented to determine the individual strengths of the different electro-tactile stimulations. The time required for this process could be reduced to about one third of the time needed for a previous work. When calibrating a certain stimulation, it had to be played 3.2 times on average instead of 11.1 times. In the first study, the 17 participants were asked to assign terms, which correspond to one sensation each, to different stimulations. The stimulation associated with Prodding was best recognized (87 %). Furthermore, the majority of the participants would also use the stimulations Vibrating, Jabbing, Prodding and Pulsating in everyday life. In total, 76 % of the participants would use the designed system in everyday life. The second study, which included a distraction task, was planned but only conducted on one expert user. An average detection rate of 88 % was achieved. This result cannot be generalized, but shows the potential of the feedback method.

The aim of this thesis is to provide an overview of recent research on Context-Awareness. Therefore, previous research dealing with conceptual modelling and practical implementation should be elaborated. In addition, the limits of context awareness in current applications should be shown. This work is suited for students who are interested in current research and not in a technical implementation. Good English skills and a high degree of self-motivation are highly recommended to explore the original English articles in detail.

The purpose of this thesis is a literature review of current research and recent advances in tactile feedback, its biological background, sensations, actuators, and use cases. The review should provide a broad background of related work and is suited for students who wish to work on a bachelor thesis which is more research-oriented and less technical. This thesis is primarily suitable for students with strong English skills and a high degree of self-motivation to explore the subject in detail, because it involves reading and reporting about original articles written in English.

A literature review of current research and recent advances in audio event (speech sounds and natural sounds) classification using a variety of different artificial intelligence approaches and their use cases is the purpose of this thesis. The review should provide a broad background of related work and is suited for students who wish to work on a bachelor thesis which is more research-oriented and less technical. This thesis is primarily suitable for students with strong English skills and a high degree of self-motivation to explore the subject in detail, because it involves reading and reporting about original articles written in English.