Dynamic gesture tracking and recognition algorithm based on deep learning

Vision-based dynamic gesture tracking and recognition has always been a difficult point in the field of computer vision. It involves multiple vision-related technologies and is an intersecting topic in multiple visual directions. The most critical dynamic gesture tracking technology and modeling recognition technology are still not mature and perfect. At present, dynamic gesture tracking technology is in the stage of transitioning from traditional tracking methods to deep learning tracking frameworks. The existing deep learning tracking frameworks are not well applicable to the visual target tracking with high degree of freedom and small space ratio such as gestures. On the other hand, it is difficult to establish an effective general gesture model for recognition due to the flexibility of dynamic gestures and complex motion processes. These problems propose a dynamic gesture recognition framework based on improved real-time multi-domain convolutional neural network tracking. The real-time multi-domain convolutional neural network is introduced to realize real-time gesture tracking, and an attention module is added to calibrate the spatial importance of features, which can clarify the discriminative area of the tracking target, and greatly improve the accuracy of gesture tracking. On this basis, the characteristics of gestures are fully considered, and the motion information, appearance information and time information of the gestures are used to establish the frame difference spatiotemporal graph model of the gestures. Finally, the deep learning network is used to recognize dynamic gestures. The CGD International Public Library of Gestures has achieved better recognition results, which offers a workable solution for dynamic gesture tracking and recognition in complex scenarios.