Research And Implementation Of Pedestrian Segmentation Algorithm On Vehicle Video

For a long time, it’s self-evident that road traffic accident has brought harm to humans, so how to reduce the traffic accidents, protect the pedestrians safety, warn drivers timely would collide the pedestrian neighbored vehicles, becomes peoples attention naturally. The research and development of pedestrian detection technology which belongs to DAS (Driver Assistance System) technology is effective way to solve this problem, moreover, pedestrian segmentation is foundation of pedestrian detection technology, segmenting pedestrians which are in video images rapidly and accurately makes for subsequent pedestrian recognition and tracking. Therefore finding a useful segmentation algorithm has potential economic value and bright future.It’s restricted the veracity and instantaneity of pedestrian segmentation that is complex background, the diversity of moving objects (including rigid people, moving vehicles and animals and so on), blocking out that is aroused by pedestrians or vehicles, change of weather and lighter and other factors. What’s more, pedestrians are moving or static on roads, so a novel algorithm for two kinds of people segmentation is needed. These are researching key in the thesis.In the thesis, for the moving pedestrians who have bigger deformation, according to the arms and legs of pedestrians have more obvious changes in successive two frames images than other parties of body, we propose image edge matching method to find quickly the mismatching features on pedestrians, and then ensure humans position.For the moving pedestrians who are not segmented, the thesis adopts global motion estimation based on Sift key-points method. When this method deal with images, we don’t consider the detected area of pedestrians who have bigger deformation any more, therefore greatly reduce processing time, achieve requirement of real-time. Through experiment of detecting various scenes of pedestrians, it’s results shows that the combine of two methods is feasible.For the static people in images, the thesis use the features that pedestrians have strong symmetry, adopt segmentation algorithm based on shape information, achieve the purpose of static pedestrian segmentation. The experiment results show that this algorithm is effective.