Research On Pedestrians Detection Based On Multi-sensors Fusion

With the increase of vehicle population in China, the traffic environmentdeteriorated, and the condition of road safety was facing severe problem. Thevulnerable groups, such as pedestrian, have suffered severe injure in traffic accidents,due to mixed traffic. Therefore, issues of pedestrian safety were urgent and important.At present, pedestrian safety protection mechanism was categorized into two classes,active safety and passive safety. Passive safety protection mechanism, in some extent,has reduced the severity of harm for pedestrian in traffic accidents, but it could notprevent the occurrence of crash. Consequently, study on effective active safetyprotection for pedestrian was urgent and necessary, which played a vital role inpedestrian life guarantee.The base of active safety protection was pedestrian detection, and now therelated algorithms have adopted single sensor. It has been found that algorithms withsingle sensor have made great effect on the whole system, due to the accumulation ofdetection errors. In this paper, multi-source fusion technology was adopted, and ahardware system for pedestrian detection was constructed based on multi-sourcefusion, such as, radar, color camera, and infrared devices, etc. The method hasavoided the limitation of information collection in the aforesaid algorithms withsingle sensor under complicated environment, improved the reliability of detectionresults, and achieved all time pedestrian detection.Key findings from the studies described in this paper were as follows:A vehicle hardware system with multi-sensors fusion was constructed.Fusion technique was referred to optimizing multi-source information,which was applied widely. Based on the purpose of pedestrian detectiononboard, we compared the advantages and disadvantages of the various sensors,a vehicle hardware system was constructed, with consideration of theprocessing speed and convenience.Pedestrian candidate regions extraction based on radar information.Based on the real-time requirements, in the system, the world coordinates ofthe obstacle in front of the vehicle were obtained from the processing of radar data, and they were transformed into image pixel coordinates throughmapping model. The effective pedestrian zone was estimated through aseries of constraints. And in the effective zone, pedestrian detection wascarried out on line, which guaranteed the precision of detection anddetection speed satisfied the real time detection.Pedestrian detection on board based on multi-source fusion wasachieved. In this paper, the pedestrian detection algorithm based on HOGfeature was optimized, and features exaction based on interesting zone andintegral histogram was introduced, for quick identification. Aiming at trafficcondition at home, pedestrian positive and negative samples in roadbackground were selected to fulfill a sample data base for various roadconditions, whose base was INRIAPerson sample data base. Then pedestrianclassifier was obtained from SVM classifier, and it showed goodresults.Based on multi-source fusion, pedestrian detection system on boardwas created from the aforesaid pedestrian detection algorithm, using VC andOpencv technique, and the system has balanced the precision of algorithmand the elapsed time.In order to ensure the reliability and instantaneityof pedestrian detection systemon board in this paper, massive tests have been applied and the results have beencompared with artificial statistics method, the new method has shown betterapplicability.