Shape Based People Detection for Visual Surveillance Systems

People detection in outdoor environments is one of the most important problems in the context of video surveillance. In this work we propose an example-based learning technique to detect people in dynamic scenes. A classification based on people shape and not on image content has been applied. First, motion information and background subtraction have been used for highlighting objects of interest, then geometric and statistical information have been extracted from horizontal and vertical projections of detected objects to represent people shape. Finally, a supervised three layer neural network has been used to properly classify objects. Experiments have been performed on real image sequences acquired in a parking area. The results have shown that the proposed method is robust, reliable, fast and it can be easily adapted for the detection of any other moving object in the scene.