With people’s attention to physical health and health preservation, more and more people practice Taijiquan to achieve the purpose of physical fitness and self-cultivation. However, many people do not fully understand the principles of Taijiquan in the early stage of practicing Taijiquan, rely on imitation to practice, and ignore the action essentials and characteristics of Taijiquan, resulting in joint pain. At the same time, many beginners ignore the causes of these situations, resulting in joint damage. Therefore, this paper proposes the research on the joint injury of Taijiquan based on computer image analysis and carries out the dependent research and analysis through the joint image processing model based on the improved Robert’s edge detection algorithm. The experimental results show that the action cycle time of wrong action is longer than the total time of correct action due to the increase of action displacement, and the long displacement distance increases the pressure on the knee joint and improves the probability of knee joint damage. In addition, there are obvious differences between wrong action and correct action in knee adduction and abduction, flexion and extension, internal rotation, and external angle. Wrong action knee joint movement does not conform to the natural movement law of human body, which increases the pressure or tension to the knee joint and surrounding related tissues and improves the probability of knee joint damage.