FPGA implementation of real-time head-shoulder detection using local binary patterns, SVM and foreground object detection

Pedestrian detection is an important feature in an advanced, automated video surveillance system. Unfortunately in most situations cameras are mounted in a way that, due to perspective, walking humans are occluded by each other or stationary objects and detecting a whole silhouette is not possible. But heads and shoulders are not occluded in most cases and can be used for object classification (human or not human) or for pedestrian counting. In the article a system implemented in FPGA for head-shoulder detection is presented. It is based on Local Binary Patterns for feature extraction and Support Vector Machines for classification. To reduce the false positives rate, foreground object detection is used as an additional validation criteria. The final system was implemented in a Xilinx Virtex 6 FPGA and is able to process a video stream of resolution 640×480@60 fps in real time.