This paper uses rehabilitation training monitoring technology based on computer image technology to monitor the exercise intensity and effect of patients in rehabilitation training in real-time. The dynamic video of the patient is captured by a high-definition camera, and the Open Pose algorithm is used to detect the key points of human movement to obtain movement data such as joint displacement and speed. The random forest model uses an integrated learning method to complete the classification of exercise intensity by constructing multiple decision trees. At the same time, ResNet (Residual Network) is applied to extract the spatiotemporal features of the movement pattern using multi-layer convolution operations, and classify and evaluate the exercise effect. Finally, the system feeds back the monitoring results to the rehabilitation trainer through a visual interface, uses metabolic equivalent of task (MET) and heart rate changes as the core indicators for exercise intensity evaluation, and comprehensively evaluates the recovery situation based on the range of joint movement, gait stability and cycle training results. The experimental results show that after 7 cycles of rehabilitation training, the patient’s recovery rate increases to 93.2%.