Abstract

In this article we propose to detect the hand and face gestures in a peer to peer video call. First the frames are captured from webcam using openCV, then hand and facial landmarks in video frames are detected using Media Pipe Holistic model. Then the extracted key point values from frames are processed and fed to the trained LSTM model which classifies the gestures into specific texts. The unique feature added is detects accurate text based on highest probability value. Then this is streamed in a peer to peer video call using WebRtc which uses signalling to establish connection among peers using http and SDP protocol.