利用邊界資訊及幾何關係之車輛後車窗邊界偵測

Rear Window Detection for On-Road Vehicles Based on the Information of Edge and Geometry Relationship

鄭兆良(Chao-Liang Cheng)

指導教授 ： 連豊力

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2014年)

https://doi.org/10.6342/NTU.2014.02500

摘要

交通影像的處理可以提供許多資訊以提供交通上的規畫以及控制，例如車輛的偵測以及可行走駕駛區域偵測提供自動駕駛系統有關附近車況的資訊，以及道路車流量的規劃。
在這篇論文中處理的對象利用紅外線閃光燈在高乘載到路旁所得到的影像。所要偵測的物體為車子的後車窗。而物體偵測通常會採用機器學習的方法，針對該物體的某些特徵訓練分辨器並辨識影像中的物體。然而，使用這種方法，辨識的好壞取決於訓練分辨器的影像，由於無法窮舉所有可能的車窗影像讓分辨器學習，所以此種方法並不實際。
在後車窗偵測上，將利用影像上線條的資訊選取可能為後車窗的結構線。利用邊緣偵測將影像轉為灰階影像，並計算影像中的消失點座標，將影像的上邊緣點及左邊緣點連線。在理論上，車窗的結構線將通過較多的邊界點，在灰階影像的白色點。透過計算每條線通過的邊界強度可以推測出哪四條線為車窗的結構線。
而得到的結構線，利用幾何關係和通過的邊界強度關係，再次確認得到的車後窗結構線是否為真正的車窗。幾何關係判斷該組結構線是否不合一般車窗的幾何形狀，或者在影像中不合理的位置。邊界強度關係判斷該組結構線是否通過一定數量的邊界強度。達到以上兩組判斷機制的結構線才可配認定為車窗。

關鍵字

交通影像 ； 結構線 ； 邊緣偵測 ； 邊界強度 ； 判斷機制

並列摘要

The processing of traffic image can provide useful information about traffic management. For example, on-road vehicle detection and drivable region detection can provide the information about nearby traffic condition to the auto-driving system and the management of traffic flow.
In this thesis, the processed objects are the images photographed by infrared flash which is next to the high occupancy vehicle lanes. The rear windows of on-road vehicles are the detection objects. The detection of objects usually uses the methods based on machine learning. The usage of features of objects is that they train the classifier to identify the objects in the image. However, the detection results are dependent on the training images for classifier. Because it is impossible to supply all kinds of images in different conditions to train the classifier, the machine learning method is impractical.
In this thesis, the rear window detection uses the line information in the image to select the possible structure lines of rear window. Edge detectors change the input images to grey level images. Then the algorithm in this thesis calculates the vanishing points’ coordinate and connects the upper side and left side points to the vanishing points. The structure lines of rear window pass the number of edge points more than other lines. By calculating the edge magnitude which each line pass, it is possible to select the candidates which are the structure lines of rear window.
The set of structure lines are judged by the geometry and edge magnitude relationship. Geometry relationship judges the set of structure lines whether is in accordance with the common shape of rear window or not. Edge magnitude relationship decides whether the set of structure lines which passed the edge magnitude exceeding a threshold. The set of structure lines which satisfy the two kind of relationship is considered as the rear window of on-road vehicle.

並列關鍵字

traffic image ； structure line ； edge detection ； edge magnitude ； judge function

參考文獻

[1: Sun et al. 2006]

[2: Foucher et al. 2011]

Philippe Foucher, Yazid Sebsadji, Jean-Philippe Tarel, Pierre Charbonnier, and Philippe Nicolle, “Detection and Recognition of Urban Road Markings Using Images,” in Proceedings of IEEE International Conference in Intelligent Transportation Systems, Washington, DC, USA, pp. 1747-1752, October 5-7, 2011.

[3: Wang & Lien 2008]

[4: Sivaraman & Trivedi 2010]