Real-time dynamic gesture track recognition method based on regression detection

The invention relates to a real-time dynamic gesture track recognition method based on regression detection. The method comprises steps of obtaining a video image, segmenting a hand region, obtainingkey points, and constructing a gesture track based on the key points; carrying out feature extraction on the information of the gesture track by using an improved TextCNN based on causal convolution and hole convolution; acquiring a feature sequence of a track, inputting the feature sequence into an LSTM, carrying out regression prediction on the feature sequence based on the changed LSTM with animproved output layer structure, clustering and screening out a most credible result from prediction results, and positioning and identifying an effective part existing in the track. The invention aims to solve a problem that a track is difficult to accurately and quickly recognize in a noise environment in dynamic gesture recognition, a method based on regression detection is provided, a regression detection algorithm of