| Platform | Hardware | Dependencies |
|---|---|---|
| 🔳 Linux | 🔳 Azure Kinect | 🔳 gflags |
| 🔳 glog | ||
| 🔳 Azure Kinect SDK | ||
| 🔳 opencv | ||
| 🔳 Anaconda | ||
| 🔳 Yolov5 | ||
| 🔳 Image annotation tool |
This project is made up of two sub-projects: image-capturing model-training. image-capturing is a CMake project that uses Microsoft's Azure Kinect to capture so-called depth color images (of cause, this can be changed). model-training uses shell and python scripts to exploit Tensor Flow's object detection API and train an object detection.
The notebooks in model-training README.md are self-documenting, but more on that in the model-training README.md. In principle, one can use any other camera, a webcam, or even already captured images (i.e., given a reasonable number of image captures exist) to train the detection model.