With the aging of the population and the continuous increase in the number of people with upper limb impairments, the rehabilitation process requires long-term treatment and supervision by professionals, making the rational allocation of medical resources a significant challenge. There is an urgent need to combine the clinical experience of physicians and the high-precision execution capabilities of rehabilitation robots to construct a upper limb rehabilitation trajectory demonstration and execution system. However, traditional drag demonstration cannot achieve non-contact demonstration and lacks flexibility, and the establishment and maintenance costs of the upper limb exoskeleton mirror system are high. In response to these issues, a non-contact upper limb terminal rehabilitation trajectory demonstration and execution system is designed using computer vision and end-effector robotic arms. Experimental results show that the constructed system can complete the “demonstration-planning-execution” human–machine interactive upper limb rehabilitation training task. It can monitor the safety in real time and optimize the trajectory deployment operation, can be used to establish a professionally effective rehabilitation action library, which helps physicians improve the training plan by adjusting the trajectory according to the patient’s condition and progress, achieve better rehabilitation effects.