Cardiovascular disease and muscle damage caused by human sports injuries are increasing. Therefore, it is of great research significance to greatly reduce the damage to the human body through sports injury monitoring, detecting the degree of injury, and taking effective measures in time. In this paper, we propose a sports injury detection mechanism based on multi-sensor fusion. First, we determine the motion area of the human body by computing the cumulative difference between frames. Second, we utilise wavelet method and AutoEncoder model to design two feature extraction methods. Third, we combine the extracted features to detect the sports injuries. The experimental results verify the effectiveness of our proposed method.