Human following serves an vital human-robotics interplay function, whereas real-world situations make it difficult notably for a cell agent. The principle problem is that when a cell agent attempt to find and comply with a focused individual, this individual may be in a crowd, be occluded by different individuals, and/or be dealing with (partially) away from the cell agent. To deal with the problem, we current a novel individual re-identification module, which accommodates three elements: 1) a 360-degree visible registration course of, 2) a neural-based individual re-identification mechanism by a number of physique elements – human faces and torsos, and three) a movement mannequin that data human’s movement and predicts human’s future place. Along with the individual re-idenfication module, our human-following system additionally tackles different challenges, reminiscent of 1) the focused individual may be fast-moving (the necessity of the system operating at a low latency for individual identification), 2) the focused individual can transfer out of the digicam sight (the necessity of trying to find the individual with out sight), and three) collision avoidance (the necessity of avoiding hitting obstacles). By in depth experiments, we observe that our proposed individual re-identification module enormously enhance human-following function when in comparison with different baseline variants.