Image-based trajectory tracking and distance estimation of a follower robot from a leader one

This article discusses a tracking algorithm applied to a multi-agent robotics system. The considering system consists of two robots: the leader and the follower. The main task is to track the leader robot with a camera, estimating the distance between the robots. Traditionally, this task requires the integration of several sensors: LiDAR, encoders, GPS. However, this work suggests a new solution in case when only a camera is available. The leader robot is captured using the background subtraction algorithm to detect the center point position of the moving object. The center point coordinates serve as input to a Kalman filter that predicts the next position of the leader robot. Simulations are conducted in the CoppeliaSim working environment where the path of the follower robot is monitored and compared to a leader robot. Moreover, two test modes are performed: the Center-Point position prediction mode and the blind tracking mode. The blind tracking mode corresponds to the case when the leader is not visible (not in the range of vision). The distance between the leader and the follower robots is estimated using the Perspective-n-points (PnP) and RANSAC (Random sample consensus) algorithm, and the results are compared with those conducted using an approximate sensor.