Analyze The Object On Image Using Computer Vision Technologies

For our scope, we need a solution to detect the face in real time on the webrtc streaming. Some state-of-the-art face detection models recently show us the very good performance, but about the inference speed, we also want real-time it too so we have researched another solution and found a very good model architecture from this paper BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs ( https://arxiv.org/pdf/1907.05047.pdf ). So we have applied this pre-trained model to our system and we see that it could handle the real-time face detection on the camera streaming with high performance as well as fast inference speed.

It's also similar to the Face detection model, we also have to face up to the performance and the inference speed problems when processing on the real-time camera streaming. So we have applied the same idea with the face detection task for the hand detection model, BlazeFace but for the hand dataset. After getting the hand location on the image, we will apply a model from this paper Attention Mesh: High-fidelity Face Mesh Prediction in Real-time ( https://arxiv.org/pdf/2006.10962.pdf ) to estimate the Hand 3d keypoints.