基于逆透视变换的车辆排队长度检测方法

Vehicles Queue Length Detection Based on Inverse Perspective Mapping and its Hardware Implementation

摘要:

交通路口的车辆排队长度检测是智能交通系统的重要组成部分,传统的检测方法易受背景噪声、摄像机透视效果等因素的干扰造成检测失败,而且其实现都是基于串行结构的处理器,不能用于实时处理的场合。本文设计了一种充分利用平直道路几何特征并适合FPGA实现的排队长度自动检测算法,该算法利用逆透视变换消除图像几何失真,引入公路的结构性约束有效检测了车道线；接着采用Sobel边缘算子检测出各车道的车辆轮廓,通过一种基于信息量的度量方法提取排队的队尾,从而确定了车辆排队长度,并且通过硬件化设计使得整个检测过程达到实时的处理速度。

Abstract:

Vehicles queue detection in intersection is an important component of the intelligent transportation system. The traditional detection methods are vulnerable to background noise and perspective effect of traffic image, and their implementation are based on the structure of serial processor, which is hard to meet the real-time processing requirements of visual applications. Thus, take advantage of the geometry characteristics of flat road, this paper develops a novel vehicles queue detection algorithm suitable for FPGA implementation. It firstly eliminates the geometric distortion of image sequence using inverse perspective mapping method, and detects the lane markers by introducing in structural constraints of road; on this basis, it extracts the contours of vehicles queue using Sobel algorithm and determines the queue tail by adopting a kind of measurement based on entropy. Moreover, the hardware architecture design in FPGA makes the entire algorithm achieve real-time processing speed. #$NLKeywords: Inverse Perspective Mapping (IPM); field-programmable gate array (FPGA); lane markers identification; vehicles queue; Sobel operator