Design of a Tennis Training Assistance System based on Sensory Recognition Technology

In order to solve the problem of traditional tennis training methods heavily relying on manual guidance from coaches (purpose), the author proposes a tennis training assistance system based on somatosensory recognition technology. The author applies somatosensory recognition technology to the design of a tennis training assistance system to achieve intelligent tennis training. The tennis training assistance system uses Kinect V2 as a real-time motion acquisition sensor and uses a separation strategy to separate characters from the background of the sports venue, thereby reducing the computational complexity of the data. Simplify the human body into 18 skeletal joints to reduce the complexity of motion recognition, and use multi-objective tracking algorithms to capture joint position data. Based on VGG convolutional neural network, 2D joint data is transformed into pose mapping, and poses at multiple time points are used as training samples, train it through guided learning methods and optimize its parameters, ultimately obtaining a real-time behavior recognition model suitable for tennis sports. By conducting 100 sets of 4 movements tests on 10 subjects, they were divided into two parts: 10 groups and 28:2. As the number of training sessions increases, the accuracy curve between the test data and training results tends to 0.86. The system design method proposed by the author is robust and practical.