A Pedestrian Detection Systembased on Thermopileand Radar SensorDataFusion
暂无分享,去创建一个
Automotive pedestrian protection systems will be introduced intheEU inshort termtoreduce thenumberof accidents andinjury fatalities. Aswithanysafety issue, acom- prehensive approach comprising bothactive andpassive safety elements should befollowed. Thisisalsovalid forpedestrian protection, whereithasbeenshownthat nexttopurely passive measures, accident avoidance systems e.g. theBrakeAssist have significant potential toreduce injury severity. Passive safety short termsolutions canbecontact sensor systems thattrigger raisable engine hoods. However, animportant enabler forafuture pedestrian protection system isasuitable, low-cost, environment- friendly sensing technology forpedestrian detection, supported byafast andreliable algorithm forobject localization. Thispaperdiscusses suchaninnovative approach forpedestrian detection andlocalization, bypresenting asystem based ontwo short rangeradars andanarray ofpassive infrared thermopile sensors, aided withprobabilistic techniques fordetection im- provement. Thetwoshort rangeradars areintegrated inthefront bumper ofthetest vehicle. Theyareable toobserve andtrack multiple targets intheregion ofinterest. However, onedifficulty isto distinguish between pedestrians andother objects. Therefore, a second sensor system isrequired toclassify pedestrians reliably. Thissystem consists ofspatial distributed thermopile sensors whichmeasure theobject presence within their respective field- of-view independently. Thesemeasurements arethenvalidated andfused using a mathematical framework. Thermopiles are excellent todetect thethermal radiation emitted byevery human. However, arobust signal-interpretation algorithm ismandatory. Inthis workastatistical approach combining Dempster-Shafer theory withoccupancy-grid methodisusedtoachieve reliable pedestrian detection. Thermopile andradarsensors useindependent signature- generation phenomena todevelop information about theidentity ofobjects within thefield ofview. Theyderive object signatures fromdifferent physical processes andgenerally donotcausea false alarmonthesameartifacts. Theintegration ofthesensor readings fromtheradarandthermopile system isconducted using aunifying sensor-level fusion architecture.
[1] Jürgen Schürmann,et al. Pattern classification , 1996 .
[2] Nico Kaempchen,et al. Sensor Fusion for Multiple Automotive Active Safety and Comfort Applications , 2004 .