Video surveillance systems are widely employed in diverse areas such as protection of vital national and local infrastructures, law enforcement, and traffic control. In typical surveillance systems, either human operators monitor activities or recorded video streams are analyzed. Smart video surveillance systems are expected to automatically analyze video data in real-time so that timely intervention may be possible. A smart video surveillance system may consist of cameras placed on stationary or moving platforms and they automatically detect, identify, and track targets of interests within the scene. This thesis focuses on development of hybrid algorithms for detection of moving targets using optical cameras that may be placed on stationary or moving platforms. Several different approaches have been implemented for target detection from static cameras. They include adaptive background subtraction, statistical background modeling, temporal filtering, and optical flow techniques. Moving object detection (foreground subtraction) algorithms typically suppress the background in the video streams by adaptive and accurate background modeling. When the camera is placed on a moving platform, the whole background of the scene appears to be moving and the actual motion of the targets must be distinguished from the background motion (global motion). The approach is to model the image motion induced by the moving platform and then remove this motion by warping the image with the inverse transformation. The image motion is modeled by parametric 2D affine transformation, which is suitable since the images are captured by the same camera in very close proximity. The detection system has been successfully implemented and tested using Vivid Datasets provided by the Air Force Research Laboratory.
展开▼
