以單眼視覺演算實現倒車障礙物偵測與距離估測

Monocular Vision-based Obstacle Detection and Distance Estimation for Reverse Parking

李維民(Wei-Min Lee)

指導教授 ： 蕭得聖

國立交通大學/電機學院/電機學院碩士在職專班電機與控制組/碩士(2009年)

https://doi.org/10.6842/NCTU.2009.01042

摘要

依據Strategy Analytics的估算，2006年全球停車安全輔助系統市場規模約為6.19億美元，估計至2013年時市場規模可望達到10.22億美元。目前市售車針對停車輔助系統，均利用單顆CCD/CMOS廣角相機，在考量不改變原有車上裝配，我們即以單顆CCD/CMOS廣角相機發展障礙物偵測與距離估測演算法。
障礙物之定義為：當駕駛者所駕駛之車輛，發生碰撞時，會造成車體損傷與生命威脅之物體皆定義為障礙物。但障礙物種類繁多，倒車環境背景複雜，所以必須要有適當特徵來描述障礙物，因此利用大部份障礙物皆有的底部陰影，與物體邊緣特性，作為是否有障礙物存在依據。依據此方法實現之障礙物偵測，可偵測至少達到4種以上障礙物。
距離估測是利用透視投影、座標平移與旋轉技巧，建立二維地面座標與二維影像座標間之關係，並利用魚眼數學模型，建立二維影像座標與二維失真影像(整數)座標關係。關係建立正確後，即可利用二維的失真影像座標求得二維地面座標的距離值。依據實驗結果，距離估測精確度可達公分等級，且誤差小於十公分。

關鍵字

障礙物偵測 ； 距離估測 ； 倒車 ； 透視投影

並列摘要

According to the report of Strategy Analytics in 2006, there is an amount of US$619-million market for worldwide parking safety assistance systems. It is estimated that the amount will rise to US$1.022 billion in 2013. The cars currently available in the market are equipped with a monocular CCD/CMOS wide angle camera for their parking assistance system. To comply with current vehicular specifications, this configuration is also adapter in this study.
Obstacles are defined as items that could cause car damage and/or human injuries in collisions. However, there are many types of complex obstacles and settings where reverse parking occurs, so a sufficiently large number of categories are required when describing these obstacles. Thus, the shadow below the obstacles and their edges could be used to detect the presence of obstacles. Using this method, one could detect more than 4 types of obstacles at a time.
Distance estimation uses perspective projection,coordinate shifting and rotating techniques to establish the relationship between the 2D ground coordinates and the 2D image coordinates, and uses a fish-eye mathematical mode to establish the relationship between the 2D image coordinates and the 2D lossy images (whole numbers) coordinates. After confirming their relationship, the distance value of the 2D lossy image coordinates and the 2D ground coordinates could be used. According to the results of the study, the accuracy rate of the distance estimation could be as close as a few centimeters, and the deviation range could be as small as 10 centimeter.

並列關鍵字

Obstacle Detection ； Distance Estimation ； Reverse Parking ； perspective projection

參考文獻

[1] Alonso.J.D., Vidal.E.R, Rotter.A, Muhlenberg.M., “Lane-Change Decision Aid System Based on Motion-Driven Vehicle Tracking.“ Volume 57, Issue 5, Sept. 2008 Page(s):2736 - 2746

[4] Kantawong. S, “Road traffic signs detection and classification for blind man navigation system.“ , Control, Automation and Systems, 2007. ICCAS '07. International Conference on 17-20 Oct. 2007 Page(s):847 - 852

[5] de la Escalera. A, Moreno. L.E, Salichs. M.A, Armingol. J.M, “Road traffic sign detection and classification.“ , Industrial Electronics, IEEE Transactions on Volume 44, Issue 6, Dec. 1997 Page(s):848 - 859

[7] Leibe.B, Schindler.K, Cornelis.N, Van Gool.L, “Coupled Object Detection and Tracking from Static Cameras and Moving Vehicles.“, Pattern Analysis and Machine Intelligence, IEEE Transactions on Volume 30, Issue 10, Oct. 2008 Page(s):1683 – 1698

[8] Takayuki Naito, Toshio Ito, and Yukio Kaneda, “The Obstacle Detection Method using Optical Flow Estimation at the Edge Image,”Proc. of IEEE Intelligent Vehicles Symposium, Istanbul, Turkey, pp.817-822, June 2007