Application Of Image Processing In Automotive Driving Assistance System

While Automotive Driving Assistance Camera System becoming a sparkling spot on the current high-end car market,we could see already some functions of ADAS available in some car types,such as Bosch's Night Vision System on Benz car,the SIEMENS VDO's Camera System on BMW's,the Continental's 360° Camera Monitor System on Audi's and KOSTAL's Lane Departure Warning System on PSA and MMC etc.Generally all these camera systems use DSP or FPGA chip for real-time image data sampling.Depending on the tasks,the systems extract the useful information of interest out of the vision by image processing and analyzing.Based on these information,the corresponding decisions need to be made to warn the driver in some certain ways.This study us dedicated to provide a software algorithm to achieve real-time Traffic Sign Detection and Recognition in Driving Assistant Systems.A little different from the most known studies which are only valid on detecting triangle warning signs or/and round prohibition signs,this study is based on most of the standard traffic signs in Chinese mainland Roads/Traffic laws,algorithm is designed by targeting on detecting most types(including different colors and shapes)of China traffic signs.TSR function as a real-time application requires consideration of the speed,during the study every step tries as much as possible to keep the efficiency of the designed algorithm.Thus the study is chosen to base on HSI color space and it uses a new type method of block-based fast color segmentation,combined with color-stratifying technology and simple threshold processing,you can quickly locate the region of interest(ROI)which might contain a traffic sign in an image.(The size of ROI can be customized.)And then in this ROI area,the basic dilation process of Mathematical Morphology is carried on to obtain smooth filling connected areas for further shape recognition.Intensity Vector image masked by the final output of region of interest,the statistical features of the edge directions are extracted for further similarity measurement in content-based image retrieval(CBIR).According to preliminary experimental results,the detection accuracy of the region of interest is more than 90%.This algorithm can be extended to most of sign shapes in the involved traffic sign database.