To solve the person following problem of binocular robot in complex environment, a method of extracting multi-
feature is presented. For the stationary robot and moving target, the improved gait flow image and view are adopted to
recognize the target, and then the color template is constructed. For both the moving robot and target, the motion information
of the head and shoulder model and the color feature extracted by extended Kalman filter(EKF), the Hu moment based
head and shoulder model matching method and the adaptive kernel based CamShift algorithm are used to track the target.
The experiments show that this method avoids selecting the person manually, and tracks the person effectively in complex
environment, such as backgroud and target high similarity, occlusion and so on.