AN INTELLIGENT FRAMEWORK FOR AUTOMATED VIDEO SURVEILLANCE SYSTEMS

Abstract

Object tracking is fundamental to automated video surveillance, activity analysis and event recognition. In real-time applications only a small percentage of the system resources can be allocated for tracking, the rest being required for high-level tasks such as recognition, trajectory interpretation, and reasoning. There is a desperate need to carefully optimize the tracking algorithm to keep the computational complexity of a tracker as low as possible yet maintaining its robustness and accuracy. After successfully tracking the moving objects from one frame to another in an image sequence, the problem of understanding object behaviors from image sequences follows naturally. Behavior understanding involves the analysis and recognition of motion patterns, and the production of high-level description of actions and interactions. Behavior analysis of a moving object in an outdoor scene is an extremely difficult task in real time. In this work, we propose and implement a novel block cluster based framework that attempts to attain a light weight tracking system by reducing undesirable and redundant computations. The frames of the video are passed through a preprocessing stage which transmits only motion detected blocks to the tracking algorithm. We also propose a novel framework which greatly assists the behavior analysis in outdoor scene. Here, we track the object in real time using a static camera and a mobile robot. The static camera uses our block based tracking framework to get the path trajectory and this path is followed by the steering camera to monitor the moving object. The steering camera can be used for the object behavior understanding. We also demonstrate the manual path tracking by the mobile robot. We aim to extend system described above for network surveillance. For this to be feasible, the images acquired by the mobile camera have to be encoded for transmission over the network. We have chosen a wavelet based JPEG 2000 encoder for this task because different techniques of wavelet based moving object detection and tracking have already been proposed. We have parallelized JPEG 2000 using CUDA for speedup. iii