Path planning of autonomous mobile robots (AMRs) is one of the current mainstreams in robotics field. It helps robots automatically plan its own moving path based on the surrounding environment information during operation to avoid collision accidents. Because a person’s gaze status often influences his next trajectory, we focus on the gaze targets of a human subject in the scene. When the robot knows the pedestrian’s next behavior, it can plan its own path to avoid collisions. In our research, we based on a gaze tracking model using 360-degree images and combined with the multi-task convolutional neural networks (MTCNN) face detection algorithm to achieve the real-time detection of human gaze targets.