ISSN 1054-6618, Pattern Recognition and Image Analysis, 2018, Vol. 28, No. 3, pp. 439–449. © Pleiades Publishing, Ltd., 2018. Visual Tracking Based on Adaptive Mean Shift Multiple Appearance Models 1 Y. Dhassi a, * and A. Aarab a, ** a Laboratory of Electronics, Signals, Systems and Computers, Department of Physics Faculty of Sciences Dhar- Mahraz, Sidi Mohamed Ben Abdellah University, Fes, Morocco *e-mail: dyounes2003@gmail.com **e-mail: aarab_abdellah@yahoo.fr Abstract—To overcome the tracking issues caused by the complex environment namely, illumination varia- tion and background clutters, tracking algorithm was proposed based on multi-cues fusion to construct a robust appearance model, indeed the global motion is estimated using the H∞ filter based on the nearly con- stant velocity motion model, then the traditional Mean Shift (MS) estimate the local state associated with each sub appearance model, finally the weights of the sub appearance models are adjusted and combined to estimate the final state. The proposed method is tested on public videos that present different environment issues. Experiences and comparisons conducted show the robustness of our methods in challenging tracking conditions. Keywords: visual tracking, mean shift, interactive multiple models DOI: 10.1134/S1054661818030057 1. INTRODUCTION Object tracking is a common problem in the field of computer vision. The constant increase in the power of computers, the reduction in the cost of cameras and the increased need for video analysis have engen- dered a keen interest in object tracking algorithms. This type of treatment is today at the center of many applications multimedia in smart visual surveillance, human computer interaction, unmanned vehicles and telerobotics. The tracking corresponds to the estimation of the location of the object in each of the images of a video sequence, the camera and/or the object being able to be simultaneously in motion. The localization process is based on the recognition of the object of interest from a set of visual characteristics such as color, shape, speed, etc. There are many challenging issues, which make the development of a tracking method [1] very difficult; the major difficulty is the object appearance changing and the background confusion. Most tracking algorithm are based on a single cue to represent the target, however one cue can’t help building a robust appearance model, thus we propose in this paper to use multiple cues to build a robust and stable appearance model, therefore dealing the prob- lem of appearance variations, indeed we adapt the IMM [2] and then it combined with the Mean Shift 1 The article is published in the original. [3], named adaptive multiple model mean shift (AMM-MS) based on multiple cues, three observa- tion model are adopted, the Mean Shift stage estimate the state on the system for several sub appearance model and the modified IMM dynamically adjusts the weights of different cues. To accelerate the convergence of Mean Shift and therefore find the optimal position in a minimum number of iterations. The global motion of the target is estimated using the H∞ filter [4], this position is considered as the initial position for the local estima- tion stage. The remainder of the paper is organized as follows: A short overview of the related work is exposed in Sec- tion 2; the appearance modeling is presented in Sec- tion 3. In Section 4, the AMM-MS tracking algorithm is presented. Experimental results are shown in Sec- tion 5. Finally, we conclude the paper in Section 6. 2. RELATED WORK In last decade many visual tracking methods have been proposed, which can be categorized in two classes, generative or discriminative. The generative methods, represent the object by appearance model that can be obviously convoluted by a kernel, then the tracking process seeks the candidate whose observed appearance model is most similar to that of the tem- plate, among the popular generative methods, one can quote, kernel-based object tracking [5], particle filter [6], interactive multiple model particle filter [7], robust online appearance models for visual tracking [8]. APPLIED PROBLEMS Received March 26, 2017