Pointer App & HW:
Current status:
We have an app that plays video from a defined camera and a when the eyetracker identifies a gaze it runs object detection on that frame and reads the name of the object detected in the defined language.
- All the code is in the following repo: https://github.com/GuyBenbenisti/PointerAppAlyn
- Installations:
-
- Tobii eyetracker - https://gaming.tobii.com/getstarted/
- Visual studio – add the tobii.interaction reference through package manager. Set the manager as StartUp Project and run it.
- Anaconda: in the environment pip install
- opencv-python
- playsound
- pyttsx3
- download yolo weights file from the web - https://pjreddie.com/darknet/yolo/
- Edit the paths in the configurations file in the manager folder – app.config. you need to insert the correct anaconda path and the project's path. If there is any problem with the paths – edit it in the Harness.cs file.
- running only the python code for object detection is also an option:
image reference: **python path**\python.exe **repo path**\YOLOv3-Object-Detection-with-OpenCV\yolo.py -i **repo path**/YOLOv3-Object-Detection-with-OpenCV/temp/image.jpeg
Web cam: **python path**\python.exe **repo path**\PointerAppAlyn\YOLOv3-Object-Detection-with-OpenCV\yolo.py
Known Gaps and Future Work:
- speed optimizations – improve the running time of Tobii, UI and Yolo.
- Object detection - the number of classes in the YoloV3 object detection is currently limited to 80. Consider using a model with more classes.
- Azure cognitive services - It is possible to replace the offline translation and text to speech and the yolo object detection with online cloud Azure cognitive services
- Buttons - add their functionality
- wrap all code to some executable easy to run