The Research And Achievement Of Lane Detection And Vehicle Detection In Driver Assistance System Based On Isomerism Embeded Platform

The increase of car ownership leads to the frequent occurrence of traffic accidents.Advanced Driver Assistant System has become a hot research in the field of scientific research. Vision-based lane-line detection and vehicle detection are the core research directions in this field. However, the existing detection algorithm has obvious problems in the application of the actual lane environment: the change of the illumination is easy to interfere the detection effect of the algorithm. The lane-line detection model is not universal.The time complexity of the algorithm is large. So its application is limited to computing platform that has good processing performance but bulky and high power consumption. For the above problems, this paper improves and realizes the lane-line and vehicle detection method about the driver assistant system based on the high performance isomerism embedded platform Jetson TK1:(1) An adaptive lane detection method is given to cope the complex illumination conditions. Firstly, the image is white balance processed by gray scale method to eliminate the influence of illumination. Then the region of interest of the lane image is dynamically divided by the combination of gray level distribution and vertical gray level search. Based on the gray level histogram and lane environment assumption, the high and low threshold of the Canny algorithm is determined to realize the adaptive lane edge detection. Secondly, use the probabilistic Hough transform to detect the characteristic points of the lane, and combine with the random sampling consensus (RANSAC) algorithm and fuzzy overlapping space to accurately fit the lane-line. After that, use the Kalman filter to track the lane detection results.and correct false detection and missed detection. Finally use image processing library OpenCV4Tegra provided by Jetson TK1 to accelerate the algorithms which are Gray world,Canny edge detection and so on in this paper. The processing time of lane-line detection method less than 30ms in 640*480 size image, and achieve a real-time property.(2) This paper analyzes the target detection algorithm based on extended Haar-like feature and Adaboost cascade classifier. use OpenCV and a large-scale vehicle positive and negative sample data to train a cascade classifier. And finally use the GPU to accelerate the algorithm. While ensuring the accuracy of detection at the same time, it can guarantee real-time detection of vehicle targets on the lane.(3) Integrated lane and vehicle detection data and the establishment of lane warning model can achieve the lane departure warning and pre-driving behavior predicts. And then use the multi-threaded programming model to distribute the detection task. Finally, through the latest QT interface construction technology QT Quick, build a friendly lane detection graphical user interface, guarantee real-time feedback of the current lane information, and make a warning to the dangerous situation.This paper based on the embedded platform to achieve the visual advanced driver assistant system,it not only has better detection and real-time,but also has a certain application value.