The goal of this project is the RoboMaster AI Challenge. In the process of preparing this competition, I combined multi-disciplinary knowledge and discovered many interesting new ideas and methods. In this game, the core part is the shooting module. The robot recognizes the enemy and shoots a small ball to hit the enemy armor board to cause damage, and wins by killing all the enemy members. The main goal of the vision work is to identify the armor board of the enemy target and control the robot's gimbal for aiming and shooting. At the same time, in order to further increase the accuracy of identification and shooting and the operating efficiency of the system, some optimization algorithms and PID control algorithm are introduced. I also add ballistic analysis to improve shooting accuracy. As an iterative version of YOLOv4, YOLOv5 has been significantly improved in all aspects. First, YOLOv5 is implemented in PyTorch, which makes it easier to understand, train, and deploy models than YOLOv4 which is developed based on the DarkNet framework. Secondly, in terms of performance, thanks to the use of a variety of computer vision cutting-edge algorithms, YOLOv5 has also greatly improved its training speed and operating efficiency. The lightweight model and good operating efficiency make YOLOv5 a very good robot vision solution. In the end, the project has achieved good results. Our robot can quickly and accurately identify the target and make considerable output. If the game is successfully held, I believe it can achieve good results. At the same time, I also hope that my work can serve as a reference and a cornerstone for our school’s future students to participate in the competition.