The framework of multi-target tracking based on neural network and motion model prediction

Abstract

Multi-target tracking technology is a key problem in many application areas, including robotics, video surveillance, and autonomous driving, and its purpose is to find tracking targets that match the characteristics in a continuous image or sensing sequence information and to form a reasonable trajectory for each target. This paper proposed a method that combines the two main existing approaches for multi-target tracking by applying the Kalman filter for motion model prediction to support the neural network target tracking under poor visibility and target shield.