傅立成臺灣大學：資訊工程學研究所蔡濬帆Tsai, Jyun-FanJyun-FanTsai2007-11-262018-07-052007-11-262018-07-052006http://ntur.lib.ntu.edu.tw//handle/246246/53660Every year many people died in traffic accidents, and the main factors are improper driving behavior or inattention to sudelen change in the surrounding environment. The circumstance stimulates the development of intelligent vehicles with driver assistance systems for enhancing the driving safety and efficiency. We are going to develop computer vision technologies as the core of such assistance system. We aim to detect the current lane region and vehicles in front of the host vehicle on road. The current lane region is bounded by two lane markings nearest to the host vehicle. For lane boundary detection, we propose a linear-piecewise lane model which has the ability to approximate arbitrary curves. For finding optimal configuration of the lane model, we first extract straight line segments and then group pairs of line segments by dynamic programming technique. For vehicle detection, we build a detector that slides a window over an image and verifies if the content inside the window represents a vehicle. The verification is based on a classifier trained by AdaBoost. Besides off-line training, we design an automatic on-line updating mechanism which can automatically learn new vehicle images and update the trained classifier. We develop and test our system on a personal computer. The test data is captured in urban scenes which contains hundreds of images.List of Figures iv List of Tables vi 1 Introduction 1 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.4 System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.5 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . 4 2 Preliminary 5 2.1 Perspective Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.1.1 Camera Con‾guration . . . . . . . . . . . . . . . . . . . . . . 5 2.1.2 The Perspective Imaging Model . . . . . . . . . . . . . . . . . 6 2.2 Lane Detection Procedures - A Review . . . . . . . . . . . . . . . . . 8 2.2.1 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . 8 2.2.2 Structure Parameterization . . . . . . . . . . . . . . . . . . . 9 2.2.3 Search Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3 Vehicle Detection Procedures - A Review . . . . . . . . . . . . . . . . 10 2.3.1 Hypothesis Generation . . . . . . . . . . . . . . . . . . . . . . 11 2.3.2 Hypothesis Verification . . . . . . . . . . . . . . . . . . . . . . 11 3 Lane Detection 13 3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3.2 Problem Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 3.3 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 3.3.1 Piecewise-Linear Lane Model . . . . . . . . . . . . . . . . . . 14 3.3.2 Constraint of line segments . . . . . . . . . . . . . . . . . . . 16 3.3.3 Evaluation of a Lane Con‾guration . . . . . . . . . . . . . . . 18 3.4 Line Segment Detection . . . . . . . . . . . . . . . . . . . . . . . . . 20 3.5 Line Segment Grouping . . . . . . . . . . . . . . . . . . . . . . . . . . 21 3.5.1 The Structure of an Optimal Solution . . . . . . . . . . . . . . 21 3.5.2 Recursive Definition of an Optimal Solution . . . . . . . . . . 22 3.5.3 Finding an Optimal Solution . . . . . . . . . . . . . . . . . . . 23 3.6 Lane Boundary Tracking . . . . . . . . . . . . . . . . . . . . . . . . . 25 3.6.1 Locating Lateral Position of the Host Vehicle . . . . . . . . . 25 3.6.2 Tracking with Kalman Filter . . . . . . . . . . . . . . . . . . . 26 4 Vehicle Detection 27 4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 4.2 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 4.2.1 Configuration of a Vehicle . . . . . . . . . . . . . . . . . . . . 28 4.2.2 Vehicle Candidates . . . . . . . . . . . . . . . . . . . . . . . . 29 4.3 Learning a Classifier for Detection . . . . . . . . . . . . . . . . . . . . 29 4.3.1 Features of a Vehicle . . . . . . . . . . . . . . . . . . . . . . . 30 4.3.2 Making Training Samples . . . . . . . . . . . . . . . . . . . . 31 4.3.3 On-Line Training . . . . . . . . . . . . . . . . . . . . . . . . . 31 4.4 Multiple Vehicle Detection . . . . . . . . . . . . . . . . . . . . . . . . 32 4.5 On-Line Classifier Updating . . . . . . . . . . . . . . . . . . . . . . . 34 4.5.1 On-Line Boosting Algorithm . . . . . . . . . . . . . . . . . . . 34 4.5.2 Automaticly Labeling New Data . . . . . . . . . . . . . . . . . 37 5 Experiment 38 5.1 Environment Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 5.1.1 Platform Information . . . . . . . . . . . . . . . . . . . . . . . 38 5.1.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 38 5.2 Datasets of Road Scene . . . . . . . . . . . . . . . . . . . . . . . . . . 39 5.2.1 Testing Images . . . . . . . . . . . . . . . . . . . . . . . . . . 39 5.2.2 Training Images for Building Vehicle Classifier . . . . . . . . . 39 5.3 Lane Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 5.4 Vehicle Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 6 Conclusion 43 Reference 45970755 bytesapplication/pdfen-US電腦視覺車道偵測車輛偵測computer visionlane detectionvehicle detection[SDGs]SDG3thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53660/1/ntu-95-R93922029-1.pdf