Design And Implementation Of The Vehicle Infrared Pedestrian Detection System Based On DM6437

Affected by the unclear sight, fatigue driving and inattention, the drivers are more proneto traffic accidents at night thanduring the day. It’s great significance to research and developa nighttime pedestrian detection system that would predict potential danger on the road toreduce the accidents at night.According to the thermal imaging principle, the infrared camera is almost free from theimpact of light condition. High quality images could be captured at night. This paper used aninfrared camera to capture the images. The DSP of Davinci series was optimized deeply forvariety of digital processing work. It’s an ideal hardware platform for Pedestrian Detectionbecause of the powerful data processing capabilities.The Pedestrian Detection system was developed on the DM6437board with CCS3.3integrated development environment provided by TI. In the stage of image segmentation, thispaper used adaptive dual-threshold segmentation method and a segmentation method thatcombined the Sobel operator and vertical projection. Most of the Regions of Interest could begot. In the stage of the pedestrian features extraction, this pager combined the HOG featuresand the LBP features. The pedestrian features could be described well by the combined twokinds of features rather than only one. In order to reduce the false alarms, this paper presentedvehicle detection to reduce the false alarms produced by the vehicles in front of the camera.In order to optimize the system, a projection based on XDC tool and DSP/BIOS real timeoperating system was built according to the characteristic of the DM6437. It’s convenience toaccelerate the real time system. It’s slow for DM6437to process the floating point operationas a fixed point DSP. This paper used IQmath library to optimize the floating point operation.In the stage of image pre-processing, VLIB library was also used to accelerate the system. Inthe coding aspect, this paper tried to use software pipelining method to programming. Such asan inline function optimization, multi-loop unrolling, reducing simple function calling andless jumps.In the end, large quantities of tests were executed. Lots of videos from all seasons in theyear were tested. The test results showed that the system could meet the requirement of practical. And the system has entered into the market as product.