整合式車輛與道路偵測及距離估測

Integrated Vehicle and Lane Detection with Distance Estimation

陳昱均(Yu-Chun Chen)

指導教授 ： 賴尚宏

國立清華大學/電機資訊學院/資訊工程學系/碩士(2014年)

https://doi.org/10.6843/NTHU.2014.00484

摘要

本篇論文提出一個行車駕駛輔助系統，系統中利用明顯相關資訊的流通，以合作的方式結合了車輛偵測、道路線偵測以及距離估測。
在行車駕駛輔助系統中行車偵測是一個很重要的議題。大多數現存的方法都是建構在sliding window的方式之上。然而，這種搜尋方式通常會有計算時間的問題以及容易造成較多的false positives，這是因為此種方法需要在影像中對所有位置使用不同大小的窗格去偵測。而我們利用路上的幾何透視關係去建構了一個高效能行車偵測方法，這種方式明顯地減少搜尋範圍。在訓練的過程中，我們利用HOG-based的行車偵測方式偵測少數幾張即時的影像去找尋有可能是行車的位置並且將這些車子視為行車候選人。接著將這些包含了幾何透視資訊的結果以配對方式計算出一個線性車寬模型。在此建構了一個利用線性車寬模型去推估的adaptive scan方法，這種方式是非常有效率的行車偵測方法。
這種經由學習的線性車寬模型提供了對道路線寬度以及畫面中水平線的限制。利用這些限制，道路線的搜尋範圍可以有效的減少。我們也對於畫面中所有可能為車道的線段使用local patch鑑定的方式去加強車道偵測的可靠性。此外，我們也對於由單眼相機擷取的影像提出了一個新穎的方法去評估距離。在我們的系統中利用已知道路線標記的關係來估測相機姿勢以及那些我們系統中偵測到車輛的距離。
從真實影像的實驗結果中顯示出，我們的系統在偵測行車與道路線以及估測前車距離上是相當穩定及準確的。結果也顯示出我們所提出來方法的精確性也比過去的方法還要來的準確。

關鍵字

高級駕駛員輔助系統 ； 行車偵測 ； 道路偵測 ； 距離評估

並列摘要

This thesis proposes an Advanced Driver Assistance System (ADAS) that combines vehicle detection, lane detection, and distance estimation in a collaborative manner.
Vehicle detection is an important research problem for Advanced Driver Assistance Systems. Most existing methods are based on the sliding window search framework. However, such methods are computationally intensive and easily produce large numbers of false positives because they need to search local windows of different scales at all positions in the image. Our efficient vehicle detection approach dramatically reduces the search space based on the perspective geometry of the road. In the training phase, we locate all possible vehicle regions from several online images by using the standard HOG-based vehicle detector and treat them as vehicle candidates. Then, pairs of vehicle candidates that satisfy the projective geometry constraints are used to estimate a linear vehicle width model. Then an adaptive scan strategy based on the estimated vehicle width model is developed for efficient vehicle detection from an image.
The learned vehicle width model provides constraints on the horizon and the lane width at different locations in the image. By exploiting the above geometric constraints, the search space for lane detection can be significantly reduced. We employ local patch constraints along hypothesized lanes extracted from the image to improve the reliability of lane detection. Moreover, we propose a novel algorithm to estimate the vehicle distance from a single image captured form a monocular camera in real time. In our algorithm, we utilize lane prior information of dash lane geometry to estimate the camera pose and the distances to the detected vehicles.
Experimental results on real videos show that the proposed system is robust and accurate in terms of vehicle and lane detection as well as vehicle distance estimation from an image. We also show superior accuracies of vehicle and lane detection compared to the previous methods.

並列關鍵字

Advanced Driver Assistance System ； Vehicle Detection ； Lane Detection ； Distance Estimation

參考文獻

[1] J.Wu and X. Zhang, “A pca classifier and its application in vehicle detection,” International Joint Conference on Neural Networks, 2001.

[3] Z. Sun, G. Bebis, and R. Miller, “Quantized wavelet features and support vector machines for on-road vehicle detection,” International Conference on Control, Automation, Robotics and Vision, 2002.

[5] M. Cheon, W. Lee, C. Yoon, and M. Park, “Vision-based vehicle detection system with consideration of the detecting location,” IEEE Transactions on Intelligent Transportation Systems, vol. 13, no. 3, pp. 1243 –1252, 2012.

[6] N. Srinivasa, “Vision-based vehicle detection and tracking method for forward collision warning in automobiles,” IEEE Intelligent Vehicle Symposium, 2002.

[8] Z. Sun, G. Bebis, and R. Miller, “On-road vehicle detection: a review,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 5, pp. 694 –711, 2006.