Platform | Hardware | Dependencies |
---|---|---|
๐ณ Linux | ๐ณ Azure Kinect | ๐ณ gflags |
๐ณ glog | ||
๐ณ Azure Kinect SDK | ||
๐ณ opencv | ||
๐ณ Anaconda | ||
๐ณ Yolov5 | ||
๐ณ 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.