Dynamic Human Tracking for Mobile Robot

The movement of the robot adds great difficulty to dynamic human tracking. The traditional method of image stabilization can not remove the error caused by the movement of the human body to the stable image, resulting in low tracking accuracy and low real-time performance. This paper improves on the basis of traditional image stabilization techniques. In this paper, the least squares method is used to fit the position of the human body in the first N frames to infer the position of the human body in the (N + 1)th frame. Subsequently, this paper adopts the method of sub-regional gray projection to separate the positions of the human body in two adjacent frames of images and stabilize the images, which greatly reduces the error caused by the movement of the human body on the stable image distance. In addition, the traditional mobile body tracking method cannot solve the occlusion tracking situation of the target human body while satisfying the real-time performance. In this paper, various strategies such as camshift algorithm, particle filtering method, image stabilization, and cross-frame difference are integrated, and a dynamic evaluation strategy of tracking quality is designed. The strategy can realize the normal tracking of the moving human body in the state of dynamic robot and tracking of the target human body in the occlusion situation.