Performance Comparison of Kalman Filter and Mean Shift Algorithm for Object Tracking

Abstract

Object tracking is one of the important tasks in the field of computer vision. Some of the areas which need Visual object tracking are surveillance, automated video analysis, etc. Mean shift algorithm is one of the popular techniques for this task and is advantageous when compared to some of the other tracking methods. But this method would not be appropriate in the case of large target appearance changes and occlusion. In addition, this method fails when the object is under the action of non-linear forces like that of the gravity e.g. a ball falling under the action of gravity. Another popular method used for tracking is the one that uses Kalman filter, with measurements (often noisy) of position of object to be tracked as input to it. This paper is based on a simulative comparison of both of these algorithms which will give a proper outline of which method will be more appropriate for object tracking, given the nature of motion of object and type of surroundings. Observations based on these methods are present in the literature but there is no evidence based on implementation of these algorithms that shows a quantitative comparison of the said algorithms

Travel Time Estimation for Pedestrian with GPS Cell Phones as Probes

“…It can estimate on past, present and future time even when the nature of model is unknown [7].Kalman filter is more flexible in case of any kind of motion other than pure translational motion.Kalman filter is based on motion while some other filter like correlation filter is based on appearance. Kalman filter performs better than Means Shift algorithm under noisy atmospheric conditions [8]. Autoregressive model is fixed and can't update to the process and measurement noise intensities, resulting in performance reduction to some extent [9].…”

A Multi-Level Visual Tracking Algorithm for Autonomous Vehicles

“…The main problems of the existing GMT tracking algorithms [42], [43], [44], [45], [46], [47], [48], [49], [50] are as follows: 1) GMT lost 2) background GMTs are moving along with the GMT 3) light brightness changes on the GMT 4) noise in the image.…”