Vehicle-mounted gesture recognition method based on deep learning

The invention discloses a vehicle-mounted gesture recognition method based on deep learning, and the method comprises the steps: tracking a gesture in real time, and obtaining a gesture image of an operator; recognizing the dynamic gesture based on a Faster-RCNN (Region Convolutional Neural Network) algorithm; a gesture instruction of a user is recognized based on a three-dimensional convolutional neural network gesture optimized in combination with an attention mechanism, and the gesture instruction is transmitted to a vehicle-mounted controller to achieve the effect of controlling a vehicle. According to the method, the Faster-RCNN anchor box generation method and the ROI pooling method are used, classification and regression of time intervals of dynamic gesture actions are achieved, positioning detection of the time intervals of the dynamic gestures in an image data sequence is achieved, and compared with traditional extraction gesture recognition, the time and the position of gesture instruction generatio