A Comparison of OpenCV Algorithms for Human Tracking with a Moving Perspective Camera

Visual tracking has received much attention in recent years, especially pedestrian tracking. People tracking represents an important computer vision problem with numerous real-world applications. While significant progress has been achieved for human tracking and detection, trackers are still prone to failures and inaccuracies to master all difficult situations that may arise during the process: changes in appearance, illumination, occlusions, camera movement or cluttered background. To overcome these limitations, tracking algorithms offered by the OpenCV software library are evaluated through this paper. These trackers are fast and easy to use. However, pedestrians are particularly difficult to track with a moving camera. This paper brings a benchmark of human tracking algorithms implementations using moving camera. Here, we propose a qualitative and quantitative assessment followed by a comparison with a particle filter algorithm based on histograms of both color and texture features. Finally, in order to compare to new developed tracking algorithms in the framework of a pedestrian tracking accuracy in an unknown environment, experiments with a new available dataset validate either the reliability of OpenCV trackers or an easy-to-use particle filter.