Abstract: Exoskeleton robots, as a new type of rehabilitation equipment, have a broad application prospect in the rehabilitation training of patients with spinal cord injuries. Human gait phase recognition is a key technology to realize the movement intention recognition of wearers and accurately control exoskeleton robots. As the knee joint cannot be locked automatically in the process of swinging the leg to the ground during the rehabilitation training of patients with spinal cord injuries, a rope-drive lower limb rehabilitation training exoskeleton robot is first designed. Foot switch sensors and a knee joint encoder are used to acquire the gait information of the patient to ensure that the knee joint is locked and released in the corresponding gait phase. Based on research on human walking gait and the need to lock and release the knee during the rehabilitation training of patients with spinal cord injuries, a gait phase classification method based on information about the contact process between the foot and the ground is proposed. Because the plantar contact process of patients with spinal cord injuries is uncontrollable and unstable, a gait phase recognition method integrating information about the contact between the foot and the ground and the swing angle and speed of the knee joint is proposed. Finally, the experimental verification of the gait phase recognition method is performed using the acquired gait data of volunteers. The results show that the average gait phase recognition rate of the proposed method is 99.906%. When the foot switch sensor fails or is not triggered normally, the correct recognition rates of the locking and releasing knee phases are 94.488% and 91.853%, respectively, demonstrating the effectiveness and fault tolerance of the proposed method.