Research On Real-time Detection Technology Of Road Forward Vehicle Base On DSP

Detection of moving vehicles in front of the road is one of the key technology of the Advanced Driving Assistance System(ADAS).The technology is significance to reducing traffic accidents and improving the driving experience.At present,vehicle detection technology mainly includes: laser radar,millimeter-wave radar and machine vision-based methods.This paper chooses the method based on machine vision and a digital signal processor(DSP)is used as a core chip to realize vehicle detection and tracking,which taking into account the technical cost and difficulty.Based on the monocular visual vehicle detection technology,it is a challenge for many domestic and foreign researchers to overcome that developing a robust,real-time algorithm.Therefore,the following work about vehicle detection algorithms has been carried out :(1)A method for determining Region of Interest(ROI)based on lane lines is studied.Firstly,the lane line edge is extracted using 135°+255° Sobel operator.Secondly,the lane line based on Hough transform is extracted,which the range of angle line is [-60,-30] or [30,60].Last but not least,the lane and vanishing point determine the trapezoid ROI.The method minimizes the vehicle search range while reducing the information interference of complex road.(2)A self-adaptive vehicle shadow threshold calculation method with variable rectangle detection window is studied.According to the variation regulation of the vehicle shadow area in the image,the threshold value is obtained by using a variable rectangle detection window with an aspect ratio of 5:1 within the ROI.Finally,experiments show that using the threshold has good effect on vehicle shadow segmentation under various road conditions.(3)A quick search method for vehicle shadow lines based on vehicle width template is studied.The vehicle width template is determined according to the ratio of the width of the vehicle to the width of the lane,and the minimum mean value is determined from the bottom to the top row by row,thereby generating a vehicle bottom hatch line.This method can quickly and accurately determine the position of the vehicle.(4)Real-time vehicle detection and tracking based on DSP.On one hand,the real-time image acquisition system is designed by DM6437.On the other hand,the vehicle detection and tracking algorithm is embedded in DSP,which based on the interval frames threshold calculation method.In the end,the experimental platform of vehicle-mounted video and toy car is designed to verify the algorithm.Conclusively,the above experimental results show that the DSP detection and tracking algorithm described in this paper has the characteristics of high accuracy,robustness and good real-time performance.The accuracy rate is 96.0% and 90.1% under simple road conditions and complex road conditions,respectively.The average processing speed per second for single-lane vehicle detection and tracking can reach 23 f/s.