| Summary: | 碩士 === 國立臺北科技大學 === 機電整合研究所 === 105 === In recent years, self-driving cars is a very popular research area, and the establishment of advance self-driving system is the extend application of various functions of advance driving assist system, ACC LKA ACS etc. The establishment of advance self-driving system(ADAS) is the low levels detection algorithm. The detection algorithm performance of LIDAR sensor is very outstanding but the drawback is a frame of LADAR sensor information is great amount, it’s request more computation resources, and the price is higher. The reason why we use camera to be our detection sensor is that there are mature technical in ADAS area and low price. This paper includes two main parts, the first part is about the camera calibration researches, in orders to project the image coordinate into real world coordinate, need to calibrate the fisheye distortion and camera attitude. The second part is about the analysis of ADAS situation, includes the role of road detection algorithm in the ADAS area, the issues of variance road detection algorithm and the issues of camera sensor. In the end, this paper proposed a road detection non-road seeds selection method for Random Walker algorithm to pick the proper seeds to improve road detection performance with a high intensity sensitive feature map (HIS Map) and image entropy pyramid. Combine the road detection result and camera calibration parameter and the corresponding position of around cameras, generate a pure image based around view drivable region. Use points cloud generated by LADAR sensor to be ground truth and compare it, we can get a similar result.
|
|---|
