基於多個倒傳遞網路分類器搭配視差圖做室外自動車導航之研究

A Study on Outdoor Guidance of Autonomous Land Vehicles Using Disparity Map Coordinated with Multi-Classifiers Based on BPNN

塗韋程(Wei-Cheng Tu)

指導教授 ： 駱榮欽

國立臺北科技大學/機電學院/機電整合研究所/碩士(2008年)

https://doi.org/10.6841/NTUT.2008.00276

摘要

本論文中，我們提出一個基於多個倒傳遞網路分類器搭配視差圖做室外道路辨識，應用於雙眼立體視覺之室外自動導航車。首先用倒傳遞網路對不同種類的道路訓練成多個分類器 (例如泥土路、草地路、柏油路、水泥路) 。當圖像輸入時，我們使用數位單眼相機上加裝3D Lens的左右影像產生的視差，先計算出視差圖得知物體的相對深度與高度，來判斷出是否有障礙物。若無障礙物，則直接選擇多倒傳遞網路分類器中辨識道路最好的為結果。若有障礙物，擷取視差圖中車子前方(近的)一部份是路的平面區域與多倒傳遞網路分類器辨識是平面區域的結果做比較，計算每個比較結果的錯誤率，最後選擇錯誤率最小的為最佳分類器，即可知道目前自動導航車是在何種道路上。透過實驗測試此系統可以實際在室外道路上行走，導航策略也可成功的閃避障礙物，來證明所提方法之可行性。

關鍵字

立體視覺 ； 自動車 ； 類神經網路 ； 道路識別 ； 物體偵測

並列摘要

In the paper, we propose a new outdoor guidance system of autonomous land vehicle (ALV) using disparity map coordinated with multi-classifiers based on Back-Propagation Neural Networks (BPNN) to implement outdoor road recognition. First, we propose the different adaptive classifiers using BPNN for different kinds of roads (such as asphalt road, cement way, grass road etc.). Furthermore, we could detect the relative depth to find the obstacles by disparity map which is generated by the camera with 3D Lens. We are only concerned about the objects within a close distance from our ALV. The road-area of non-obstacle will be selected as input of each classifier. Following the results, the preferable classifier could be selected by computing the error percentage of classifier to know what kind of road where the ALV is on. Experimental results on real road scenes have substantiated the effectiveness of the proposed method.

並列關鍵字

Stereo Vision ； Autonomous Land Vehicle ； Neural networks ； Road recognition ； Object Detection

參考文獻

[1] M. Bertozzi, A. Broggi, A. Fascioli, “Vision-based intelligent vehicles: State of the art and perspective,” J. of Robotics and Autonomous Systems, pp. 1-16, June 2000.

[4] 葉哲昌，利用鬆弛相關性改進3D對應點匹配與重建應用於室外自動導航車，碩士論文，國立臺北科技大學自動化科技研究所，台北，2006。

[6] Yinghua He, Hong Wang, and Bo Zhang, “Color-Based Road Detection in Urban Traffic Scenes,” IEEE Trans, Intelligent Transportation Systems, Vol. 5, No. 4, pp. 309-318, Dec 2004.

[7] Ma, J.; Ahuja, N., “Region Correspondence by Global Configuration Matching and Progressive Delaunay Triangulation,” IEEE Conference on, Volume: 2, 13-15 June ,2000.

[8] Il-Kyun Jung; Lacroix, S., “A robust interest points matching algorithm,” Eighth IEEE International Conference on, Volume: 2, 7-14 July, 2001.