使用3D視覺資訊偵測道路和障礙物應用於人工智慧策略之室外自動車導航

Road and Obstacle Detection Using 3D Vision Information Applied to Outdoor Guidance of Autonomous Land Vehicle by Artificial Intelligent Policy

陳禹旗(Yu-Chi Chen)

指導教授 ： 駱榮欽

國立臺北科技大學/機電學院/電腦與通訊研究所/碩士(2006年)

摘要

在本篇論文中，我們提出一套以特徵點萃取、影像點對應及立體電腦視覺技術為基礎來得到前方場景的3D資訊為系統主架構。本系統主架構是利用兩個CCD所拍攝到的影像來計算目前場景的3D資訊，並藉由場景的3D資訊可以推測出目前在車子的前方是否有障礙物的存在，也可以推測出障礙物離車子的距離和方位，進而再搭配上人工智慧的導航策略，使得自動車是可以利用3D資訊來瞭解前方的環境進而做導航和避碰。
在利用立體電腦視覺開始前，攝影機校正是必要的，我們利用特定8點的3D點及投射到左右攝影機中之影像點以最小平方法來求得左右攝影機校正參數，之後我們可利用這些參數及左右影像點進行影像中景物之3D重建。由於我們利用立體視覺來取得環境的3D資訊，因此影像點對應是必要的，得到影像對應點進一歩得到對應點的3D資訊。影像對應點是一個相當重要且有待改善的問題，影像對應點將對場景的3D資訊會有著很大的影響，所以我們利用Harris Corner Detector去找出影像中較特殊的點，再利用一些幾何上的限制跟 fundamental matrix 搭配來尋找出一組最佳的對應點。在子目標特徵搜尋方面我們利用物形比對的方法，依據景物的長度、角度、面積及景物之間的間隔距離來萃取出特定景物，如軌道線及圍牆上的交叉線當作地圖，並依此做為是否到達子目標或目標位置的依據並往下一個子目標或目標前進或是下達停車指示。
當我們取得重建的3D資訊，了解車子前方場景的空間資訊後，就可得知障礙物與路面在空間中的位置並利用車子與子目標或目標之間的方向角規劃出最佳的導航路徑，使得車子可以安全繞過障礙物並在最佳路徑前提下朝向目標前進。

關鍵字

電腦視覺 ； 自動車 ； 攝影機校正 ； 特徵點萃取 ； 影像點對應 ； 3D 重建

並列摘要

In the thesis, we have developed a system to obtain 3D reconstruction of the front scene equipped with feature point extraction, stereo correspondence and binocular stereovision system. The system uses two cameras to reconstruct the 3D structure of a scene. We can utilize the 3D information of the scene to understand the environment and determine the obstacles positions and orientations. Hence, The ALV can utilize the 3D information to navigate and obstacle avoidance in the outdoor environment using AI policy.
Before using binocular stereovision system, the camera calibration is necessary. We employ the linear least-square method to obtain calibration parameters of the left and the right cameras using eight known 3D points and image points projected from real world into cameras. Then we can reconstruct the 3D information by using the calibration parameters and the image points of two cameras. Due to we utilize stereovision to know the 3D information of the scene, the correspondence problem is the important and the most difficult problem of 3D reconstruction. The accuracy of stereovision correspondence will be affected greatly the 3D information. We use Harris corner detector to extract the feature points of the images. These feature points are candidates using the fundamental matrix and geometry constraints to look for the best correspondence points. In the subgoal and the goal searching, the string match approach is employed to find out them. In the navigation and path planning, we define some features such as length, angle, area and distance of the objects as a map. Following the map, we find the subgoal and the goal. In here the subgoal is a pair of rail lines and the goal is the cross lines on the wall beside the road. When they are found out, it indicates that the ALV has arrived at the position of the subgoal or the goal and runs toward the goal or stop.
After we derive the 3D structure of a scene, we can understand the environment and determine the obstacles positions and orientation. The direction between ALV and the subgoal or the goal is obtained by an E-compass. We employ an AI-based navigation method to obtain the angle where ALV has to turn and make ALV avoid the obstacle safely and run toward the subgoal or the goal in an appropriate path.

並列關鍵字

Autonomous Land Vehicle ； Computer Vision ； Camera Calibration ； Feature Point Extraction ； Stereo Correspondence ； 3D Reconstruction.

參考文獻

[3] Zhengyou Zhang, “A Flexible New Technique for Camera Calibration,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 50, no.1, Nov. 2003, pp.1330-1334.

[4] Changming Sun, “Fast Stereo Matching Using Rectangular Subregioning and 3D Maximum-Surface Techniques,” International Journal of Computer Vision, vol. 47, no.1/2/3, May 2002.

[5] Heiko Hirschmぴuller, “Improvements in Real-Time Correlation-Based Stereo Vision,” Published in Proceedings of IEEE Workshop on Stereo and Multi-Baseline Vision, Kauai, Hawaii, Dec. 2001, pp141-148.

[8] Yajun Fang, Ichiro Masaki and Berthold Horn, “Depth-Based Target Segmentation for Intelligent Vehicles: Fusion of Radar and Binocular Stereo,” IEEE Transactions on Intelligent Transportation Systems, vol. 3, no.3, Sep. 2002.

[12] Chris Harris and Mike Stephens, “A Combined Corner and Edge Detector,” 4th Alvey Vision Conference, 1998, pp147-151.