This article discusses a tracking algorithm application in a multi-agent robotics system. This system consists of two robots, the first is the leader and the other is the follower. The main task is tracking the leader with a camera. This task can be done by using many sensors like: LiDAR, encoders, GPS. But the research is done as a solution for that situation when just a camera is available. In general, Tracking the leader aims to keep it in the range of vision and that can be done through many steps. First, the robot body should be detected using a classifier like Haar or HOG (Histogram of oriented objects), These classifiers are trained on photos for the leader in many positions, so detection can be done with help of special features in the leader. After detection, all information that is not important can be deleted which makes the process faster. Secondly, tracking can be done by subtracting the current frame from the last one, resulting in the identification of the moving object's center. The center coordinates perform an input for Unscented Kalman Filter which works as a predictor for the next step and a noise filter. The results can be tested on Vrep and the trajectories of the robots can be compared. Also, the algorithm was tested against the Gaussian noise and gave good results.