利用路面辨識資訊改進環場影像下的光流與視差法之障礙物偵測機制

行車時當駕駛者窄巷會車、左右轉彎或是停車時，常常會有擦撞或是車禍的情況發生，所以偵測出車輛周圍的障礙物，進而提升行車安全性一直是很重要的議題。光流法及視差法都被使用在環場影像下進行障礙物偵測，然而只使用這類的演算法判斷障礙物時還是會有誤辨識的情形發生，因此本論文提出利用路面辨識的結果來改進環場影像下光流與視差的資訊，能更有效的減少障礙物誤判的情形。一開始我們使用Adaboost演算法來進行路面辨識，對不同的路面種類進行分類和訓練，再來根據路面辨識的結果和先前所取得的光流點，可以更準確地計算出車輛移動資訊並且同時得到只屬於地面的光流點。並使用車輛移動資訊調整過後的前後張環場鳥瞰影像進行視差運算，找出視差與高度的關係，即可辨識出障礙物區域及其相對高度。最後結合上述資訊，顯示出更準確的障礙物區域。本研究利用CUDA平行運算,加速此系統的運行。我們所提出的方法比只使用視差法偵測障礙物時，減少平均7.49%的錯誤率。

關鍵字

鳥瞰；環場鳥瞰影像； Adaboost ；光流；視差法；障礙物偵測；影像處理

並列摘要

When drivers pass over with others in narrow alleys, park or having right or left turns, crashes and accidents often occur. As a result, it has been an important issue to detect the obstacles in the surroundings around vehicles. Under the around view monitor (AVM), optical flow and stereo disparity are used to detect the obstacles around. However, there always still be several errors existing by using only this kind method. In this paper, we proposed an improving obstacle detection mechanism under AVM based on Adaboost Algorithm combining optical flow and disparity information. So that detecting errors would be decreased. In the beginning, Adaboost Algorithm is introduced for implementing road detection, and different types of classifiers are trained to detect different road surfaces. With the results of road surface detecting, we verify the optical flow which used optical flow method only, and we get the corrected optical flow points of the ground, and meanwhile, we can figure out the movement of the vehicles. Furthermore, stereo disparity is employed to find out the relation between obstacle’s disparity and its height. Consequently, based on the improvement with the combination of ground optical flow and stereo disparity method information, we can get the region of obstacles more accurate. In this study, we also use CUDA parallel computing to speed up the calculation. Compared with using the stereo disparity method only, the proposed method leads to the decrease of the error rate by 7.49 % on average.

並列關鍵字

Bird’s-eye View ； Around View Monitor ； Adaboost ； Optical Flow ； Disparity ； Obstacle Detection ； Image Processing

參考文獻

[14]沈思瑜，改良式的多分類器路面辨識應用於遠端呼叫自走車的導航研究，碩士論文，國立臺北科技大學電腦與通訊研究所，臺北，2011。

[1]S. Shah and J. K. Aggarwal, “Intrinsic parameter calibration procedure for a (high-distortion) fish-eye lens camera with distortion model and accuracy estimation,” Pattern Recognition, vol. 29, no. 11, 1996, pp.1775 – 1788.

[2]K. Sung, J. Lee, J. An and E. Chang, “Development of image synthesis algorithm with multi-camera,” Vehicular Technology Conference (VTC Spring), 2012, pp. 1-5.

[3]B. Thomas and Y. Picard, “Development of a Cost Effective Bird’s Eye View Parking Assistance System,” Recent Advances in Intelligent Computational System(RAICS), Sep, 2011, pp.461 – 466.

[4]J. Liu, Z. Li, H. Zhang and C. Lv, “A Vision-Based Road Recognition Algorithm,” in 3rd IEEE Conference on Industrial Electronics and Applications, 2008, pp. 284-287.

國際替代計量

利用路面辨識資訊改進環場影像下的光流與視差法之障礙物偵測機制

全文下載

主題瀏覽