Abstract: To address impact of dense orchard foliage obstructing GNSS signals and affecting positioning navigation, as well as poor adaptability and limited flexibility of traditional transport robot follow-up methods, a visual tracking positioning method for transport robots was proposed．By integrating gesture recognition within YOLO11n pose estimation framework, a human keypoint recognition and localization model was constructed．Combined with a depth camera, median filtering, and quaternion methods were employed to calculate spatial coordinates and estimate pose of target individuals．To validate system performance, static hand-raising tests and dynamic recognition anti-interference tests were designed．Test results demonstrated that within a range of 1.0 to 2.0 meters from camera, average recognition accuracy exceeded 92.5% during static hand-raising tests．At a distance of 1.5 meters from camera, recognition accuracy reached 90% in dynamic recognition anti-interference test conditions, indicating robust automatic recognition and stable dynamic positioning capabilities．This technological research findings could provide valuable reference for target positioning related studies．