Real-Time Estimation of Vehicle Speed Based on Feature-Driven Multi-Frame Optical Flow
Date Issued
2004
Date
2004
Author(s)
Yen, Chia-Ho
DOI
en-US
Abstract
Traditionally, the radar-based speed estimation devices are generally used in traffic monitoring systems, but they have several disadvantages such as expensive, low estimation precision and misdetection of some vehicles. Because of these reasons, we attempt to design a vision-based vehicle speed estimation system instead of traditional radar-based ones. In this thesis, we design a real time velocity estimation traffic monitoring system based on feature-based multi-frame optical flow. Our system is composed by two parts: vehicle detection and vehicle speed estimation. . In vehicle detection, we construct a background model by assuming the pixel as mixture of Gaussian distribution, and then we can locate moving objects by subtracting each image by this model. We also use an algorithm to prevent false detection like object shadow or what caused from camera vibrating in this system. . In vehicle speed estimation stage, we proposed an algorithm to estimate optical flow across multiple frames, and find the correspondence of the moving objects in each frame and then transform it into real vehicle velocity through a known intrinsic parameter camera. Our system can achieve high accuracy of moving object detection, high reliable of flow estimation and real-time goal.
Subjects
背景模型
光流
車輛監督系統
Background Modeling
Optical Flow
Traffic Monitoring System
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-93-R91922074-1.pdf
Size
23.31 KB
Format
Adobe PDF
Checksum
(MD5):019c6ceca81b11cb61d0e3b289173cfc