Particle swarm optimization based object tracking using HOG features

Image based object tracking has always remained a challenging task because of the numerous video complexities, such as illumination variations, posture or view-angle alterations, object appearance changes, partial and full occlusions etc. Another important constraint is the necessity of real-time processing of online video stream. The tracking technique and object appearance model play a critical role in the success of a tracker. This work presents a new methodology for object tracking `IS-ObjTrack´, which utilizes a computational intelligence based tracking algorithm, employing the particle swarm optimization (PSO) technique. PSO provides robustness and time efficiency. The major advantage of the proposed IS-ObjTrack is the utilization of histogram of oriented gradients (HOG) for the development of an object appearance model. The proposed HOG based appearance model is readily exploited by PSO for fast i.e. real-time object tracking. HOG belongs to the class of gradient based filters, hence shows excellent results for objects with distinguished edges. The appearance model is designed for adaptation, whereby the parameters are updated in this work in an online manner. Experimental comparison with existing intelligent tracking systems shows the efficiency of the proposed IS-ObjTrack approach.