Design and implementation of novel image segmentation and BLOB detection algorithm for real-time video surveillance using DaVinci processor

• Published in: ICACCI : 2014 International Conference on Advances in Computing, Communications and Informatics : 24-27 September 2014 pp. 1909 - 1915
• Main Author: Patro, Badri Narayana
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2014
• Subjects: Algorithm design and analysis; Background Subtraction DaVinci Processor; BLOB; Brightness; Centroid Based; Clustering algorithms; Image segmentation; Noise; Object tracking; Optimization; Segmentation; Streaming media; Tracking
• ISBN: 1479930784; 9781479930784
• DOI: 10.1109/ICACCI.2014.6968360

A video surveillance system is primarily designed to track key objects, or people exhibiting suspicious behavior, as they move from one position to another and record it for possible future use. The critical parts of an object tracking algorithm are object segmentation, image clusters detection, and identification and tracking of these image clusters. The major roadblocks of the tracking algorithm arise due to abrupt object shape, ambiguity in number and size of objects, background and illumination changes, noise in images, contour sliding, occlusions and real-time processing. This paper will explain a solution of the object tracking problem, in 3 stages: In the first stage, design a novel object segmentation and background subtraction algorithm, These algorithm will take care of salt pepper noise, and changes in scene illumination. In the second stage, solve the abrupt object shape problems, objects size and count various objects present , using image clusters detected and identified by the BLOBs (Binary Large OBjects) in the image frame. In the third stage, design a centroid based tracking method, to improve robustness w.r.t occlusion and contour sliding. A variety of optimizations, both at algorithm level and code level, are applied to the video surveillance algorithm. At code level optimization mechanisms significantly reduce memory access, memory occupancy and improved operation execution speed. Object tracking happens in real-time consuming 30 frames per second(fps) and is robust to occlusion, contour sliding, background and illumination changes. Execution time for different blocks of this object tracking algorithm were estimated and the accuracy of the detection was verified using the debugger and the profiler, which will provided by the TI(Texas Instrument) Code Composer Studio (CCS). We demonstrate that this algorithm, with code and algorithm level optimization on TIs DaVinci multimedia processor (TMS320DM6437), provides at least two times speedup and is able to track a moving object in real-time as compared to without optimization.