Affiliations: Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran | Electrical and Computer Engineering Department, Khajeh Nasir Toosi University of Technology, Tehran, Iran
Note: [] Corresponding author. Mahsa Mirzamohammad, Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran. E-mail: Mahmir3@gmail.com
Abstract: In this paper we use cellular learning automata integrated with a normalized gradient based motion detection algorithm in fuzzy domain to detect and track the moving objects in a video. A sequence of the video frames with a preset interval are first converted to gray scale images and then based on the first frame a first order gradient is calculated in fuzzy domain. Normalization process is then performed and the gray levels are ranged between 0 and 255 for each pixel. A sequence of primary motion detected images are calculated from the normalized first order gradient images subject to a minimum threshold value for difference between every two sequent gradient images. If the condition of a minimum difference is met, each primary motion detected image is therefore computed as the products of the second order gradient of the image by the difference of the current frame with the average. Two terms of the products provide larger values for larger motions while vibrations and slight shakings are avoided. Resulting motion images are normalized to the gray scale range [0 : 255]. Then, cellular learning automata algorithm is used to reconstruct and form the object(s) of interest based on a set of rules. Objects are detected by a contour after the reconstruction. Detected objects in the sequential frames are used for the tracking purpose. This algorithm can be used for controlling and supervising a secured public place viewed by a fixed camera. The objects may be considered as the traversing people, animals, cars or any carrying baggage and the algorithm can be parameterized for the objects of interest. Finally, all parameters of the method as the threshold values, score, penalty and the number of evolution cycles are analyzed to find the optimum values for the dataset under analysis. Comprehensive experiments are performed to show the capability and efficiency of the proposed method while it is stated that developing this code in MATLAB constrains working in offline processing mode.
