The rapid development of sports training technology promotes the improvement of physical fitness of the sports population, but sports lower limb functional injuries inevitably occur. In order to realize effective lower limb rehabilitation, the recognition of rehabilitation posture for sports lower limb motor function injury becomes crucial. In this paper, we preprocess the sports lower limb movement action data through the method of error calibration and data normalization. The DeepConvLSTM neural network is proposed by combining convolutional neural network (CNN) and long short-term memory (LSTM) recurrent neural network, and the gesture recognition model based on DeepConvGRU neural network is constructed by integrating gated recurrent unit (GRU). The performance of the model on the lower limb motor function injury rehabilitation task is evaluated through experiments. The DeepConvGRU model in this paper achieved 96.83% and 99.11% accuracy on the UCISmartphone dataset and RehaLab-412 dataset, respectively, demonstrating good model performance.
