Camera Calibration and Image Tracking for Traffic Parameter Estimation
|關鍵字:||動態攝影機校正;背景建立;影子偵測;偵測窗;主動輪廓模型;卡曼濾波器;交通監控;轉彎率偵測;光流法;影像處理;影像追蹤;dynamic camera calibration;background genration;cast-shadow detection;detection window;active contour model;Kalman filter;traffic monitoring;turn ratio estimation;optical flow estimation;image processing;image tracking|
|摘要:||本論文主要探討運用於交通參數估測系統之先進影像處理技術，包括動態攝影機參數校正、背景影像建立、影子移除、車輛偵測追蹤與交通參數量測。進行交通參數估時，首先必須準確地校正攝影機參數，方能利用二維影像畫面來精確地求出物體於空間之三維物體位置，本論文設計一套可自動校正Pan-Tilt-Zoom攝影機參數之方法，發展出利用一組平行的車道標線與其在影像畫面中之幾何位置，產生焦距方程式來求焦距，再計算出其他攝影機參數值。在車輛偵測與追蹤前處理階段，一般利用背景影像移除方法將移動中車輛由影像畫面中分離出來，本論文提出基於群組直方圖 (Group-Based Histogram)方法，可快速建立良好之背景影像，此方法對於感測雜訊與慢速移動車輛具強健性。進行交通監測影像分析時，車輛的影子會造成車輛影像嚴重變形，甚至與其他車輛影像產生重疊，嚴重影響車輛偵測與追蹤的準確性。本論文利用影子色彩特性與統計方法，提出一個色彩空間比值模型，此模型可迅速偵測影像中之影子像素，配合兩種幾何分析方法再提昇影子偵測的準確性。以這個比值模型所設計的動態車輛偵測方法比現有方法更有效率。交通參數估測時，必須對多車道中大小不一之車輛進行追蹤，本論文發展一套自動輪廓初始化方法，利用特殊設計之偵測窗來偵測進入影像畫面中多車道內任意位置且大小不同之車輛，並依據車輛大小與位置所產生車輛之初始追蹤輪廓，再利用卡曼濾波器進行追蹤，分析追蹤結果可得到車流與車速之交通參數。另外，針對T字路口轉彎率量測，本論文設計一套可即時判定車輛移動方向的光流偵測技術，並結合偵測窗來量測路口之車輛轉彎率。本論文所發展之方法理論，均已利用實際道路交通影像進行驗證，量測所得之交通參數，如平均車速、車流量、車流密度與轉彎率，誤差值在5％內，顯示本論文所提出的方法確實能正確且快速完成交通參數估測。|
The objective of this thesis is to study advanced image processing methodologies for estimating traffic parameters with functional accuracy. The developed methodologies consist of camera calibration, single Gaussian background modeling and foreground segmentation, shadow suppression, vehicle detection and tracking, and optical-flow-based turn ratio measurement. The accuracy of estimating vehicle speed depends not only on image tracking but also on the accuracy of camera calibration. A novel algorithm has been proposed for automatic calibration of a pan-tilt-zoom camera overlooking a traffic scene. A focal length equation has been derived for camera calibration based on parallel lane markings. Subsequently, the pan and tilt angles of the camera can be obtained using the estimated focal length. To locate the parallel lane markings, we develop an image processing procedure. In the preprocessing step of vehicle detection and tracking algorithm, foreground segmentation can be accomplished by using background removal. The quality of background generation affects the performance of foreground segmentation. Thus, a group-based histogram algorithm has been designed and implemented for the estimation of a single Gaussian model of a background pixel in real-time. The method is effective and efficient for building the Gaussian background model from traffic image sequences. It is robust against sensing noise and slow-moving objects. However, shadows of moving objects often cause serious errors in image analysis due to the misclassification of shadows as moving objects. A shadow-region-based statistical nonparametric method has been developed to construct a RGB ratio model for shadow detection of all pixels in an image frame. This method of shadow model generation is more effective than existing methods. Additionally, two types of spatial analysis have been employed to enhance the shadow suppression performance. An automatic contour initialization procedure has been developed for image tracking of multiple vehicles based on an active contour and image measurement approach. The method has the capability to detect moving vehicles of various sizes and generate their initial contours for image tracking in a multi-lane road. The proposed method is not constrained by lane boundaries. The automatic contour initialization and tracking scheme has been tested for traffic monitoring. Additionally, this paper proposes a method for automatically estimating the vehicle turn ratio at an intersection by using techniques of detection window and optical flow measurement. Practical experimental studies using actual video clips are carried out to evaluate the performance of the proposed method. Experimental results show that the proposed scheme is very successful in estimating traffic conditions such as traffic flow rate, vehicle speeds, traffic density, and turn ratio.
|Appears in Collections:||Thesis|
Files in This Item: