Human beings are prone to limb fractures or physical sports injuries during exercise, and patients with sports injuries will no longer be followed up after treatment to relieve pain, resulting in sequelae of injuries caused by patients, and when through a complete rehabilitation assessment, not only can improve the sequelae, but also strengthen the body. However, it is very important to effectively record the patient’s rehabilitation process and provide medical staff with a complete rehabilitation record so that the physician can grasp the patient’s training results. In this paper, the node sensor is used to detect gait, which is divided into walking, running and rope skipping, and the characteristics are expressed by the pedometer calculation of the number of steps and frequency domain detection, and the similarity is obtained by using the dynamic time warp time (DTW) calculus to determine the patient’s gait abnormality through real-time comparison between the similarity and the normal gait, and the similarity of the training can reach 8%, and the error rate of the pedometer can reach 99–100%.