Based on the continuous leap of science and technology and the continuous progress of society, computer vision technology has penetrated into all walks of life, especially in the field of safety monitoring. For a long time, the safety of children in outdoor activities has been the focus of attention from all walks of life, and complex environmental factors have undoubtedly aggravated this risk challenge. Traditionally, relying on manual monitoring is inadequate because it is difficult to achieve immediate and comprehensive monitoring. This paper aims to explain how to develop a position tracking and safety risk early warning system specially designed for children’s outdoor games with the help of computer vision technology. The system deeply integrates the deep learning technology, and through the processing and depth analysis of the collected images, not only the children’s position can be accurately locked, but also their behavior patterns can be effectively identified. Furthermore, the risk assessment mechanism built into the system will objectively judge the safety of children’s behavior, and once any potential threat is detected, it will immediately trigger an early warning signal. The experimental data prove that the system shows excellent performance in children’s position tracking and safety risk early warning, which significantly reduces the situation of false alarm and missed alarm, and greatly improves the accuracy and reliability of safety monitoring.