利用函數類神經模糊網路於即時車道線偵測及前車測距之研究

Applying a Functional Neuro-Fuzzy Network to Real-Time Lane Detection and Front-Vehicle Distance Measurement

林榮華(Jung-Hua Lin)

指導教授 ： 林正堅 ； 廖梨君

朝陽科技大學/資訊學院/資訊工程系/碩士(2009年)

摘要

交通事故中發生車禍的原因，大部分來自於駕駛者的分心、疲勞或是不注意車況，而發生危險。因此，為了能盡量避免駕駛者陷入危機狀況，在本論文中我們將使用一部架設在車上的攝影機擷取前方道路的影像來偵測道路情況，並設計一套即時車道線偵測及前車測距之系統，使駕駛者能安全駕駛。本系統分為兩部分，分別為車道線偵測，以及前車測距。
在車道線偵測部份，使用遮罩找出車道線的邊緣，利用它傾斜的特性找出車道線，利用先前偵測的車道線去計算兩者之間角度對稱關係，判斷車輛是否偏移，並給予不同的偏移警告訊號。在前車測距部份，利用陰影找出車輛所在位置，透過索貝爾邊緣偵測找出車身，將車輛進行定位，接著使用函數類神經模糊網路 (functional neuro-fuzzy network, FNFN)，透過攝影機擷取前車底部陰影為輸入，輸出則為真實距離。
本論文中，我們提出的FNFN是使用一個函數鏈結類神經網路(functional link neural network, FLNN)到模糊法則的後件部，其中在FLNN的函數展開部分主要是利用直交多項式和線性獨立函數的特性。因此，在FNFN後件部分可以產生一個非線性的輸入變數組合。以此方式的設計更能增強函數逼近的準確度。在學習演算法部分，主要包含架構學習和參數學習。其中，架構學習是取決於熵值的測量來決定模糊法則的數量，而參數學習是使用倒傳遞演算法調整歸屬函數的形狀和FLNN所對應的權重值。
雖然大部份的即時前車測距系統使用攝影機座標換算成真實世界座標來表示距離；然而，從實驗的結果證明所提出的函數類神經模糊網路架構也能有效的應用於距離偵測。本文發展的系統在Intel 3.2GHz的電腦平台執行，執行速度每秒可處理高達40張影像。測試影像為台灣高速公路的實際駕駛影片，從實驗結果我們可以發現系統都維持穩定的偵測結果。

關鍵字

前車距離偵測 ； 車道線偵測 ； 函數類神經模糊網路 ； 熵 ； 倒傳遞演算法

並列摘要

Most traffic accidents result from the distracted condition, inattention to the adjacent cars, and driving fatigue of the driver. Therefore, in order to avoid the driver being in danger as much as possible, we set up a camera inside the vehicle to capture road condition and designed a real-time lane detection and front vehicle distance system for safe driving. The system is divided into two parts, the lane detection and the front-vehicle distance detection.
As for the lane detection, we used the mask to find (search for) the edges of lane lines, then identified the particular lane according to the steep degree of the lane lines. The system futures in calculating angle relations of the boundaries to determine the different degrees of the departure from the vehicle, in order to send a suitable warning signal to drivers. In the part of the front-vehicle distance detection, to locate the vehicle, shadow was utilized to find out the location of vehicle and Sobel edge detection to find out the vehicle body. We input the shadow at the bottom of the vehicle, which was captured by the camera, then output the real distance by using the functional neuro-fuzzy network.
The proposed FNFN model uses a functional link neural network (FLNN) to the consequent part of the fuzzy rules. This study uses orthogonal polynomials and linearly independent functions in a functional expansion of the FLNN. The FNFN model can generate the consequent part of a nonlinear combination of input variables. Thus, the designed can improve the accuracy of functional approximation. The learning algorithm, which consists of structure learning and parameter learning, is also presented. The structure learning depends on the entropy measure to determine the number of fuzzy rules. The parameter learning, based on the back-propagation algorithm, can adjust the shape of the membership function and the corresponding weights of the FLNN.
Although most real-time front vehicle distance systems display distance must use camera coordinates conversion to world coordinates; however, from the results of the experiments, the proposed functional neuro-fuzzy network module can be used as a feasible and effective system for distance detection, too. The real-time front vehicle distance system proposed in this thesis has been successfully evaluated on the PC platform of Intel 3.2-GHz CPU with an average frame-rate up to 40fps. Moreover, this algorithm can maintain stable results when driving on Taiwan’s highway.

並列關鍵字

front-vehicle distance measurement ； lane detection ； functional neural fuzzy network ； entropy ； back-propagation algorithm.

參考文獻

[1]Y. He, H. Wang and B. Zhang, "Color-based road detection in urban traffic scenes," IEEE Transactions on Intelligent Transportation Systems, Dec. 2004, Vol. 5, No. 4, pp. 309-318.

[2]D. J. Kang and M. H. Jung, “Road lane segmentation using dynamic programming for active safety vehicles,” Pattern Recognition Letters, 2003, Vol. l, No. 16, pp. 3177-3185.

[5]A. Bodis-Szomoru, T. Daboczi, and Z. Fazekas, “A far-range off-line camera calibration method for stereo lane detection Systems,” Instrumentation and Measurement Technology Conference Proceedings, May. 1-3, 2007, pp. 1-6.

[6]Q. B. Truong and B. R.Lee, “New lane detection algorithm for autonomous vehicles using computer vision,” International Conference on Control, Automation and Systems, Oct. 14-17, 2008, pp. 1208-1213.

[8]J. F. Liu, J. H. Wu, and Y. F. Su, “Development of an Interactive Lane Keeping Control System for Vehicle,” Vehicle Power and Propulsion Conference, Sept. 9-12, 2007, pp. 702-706.