刘派 廖寿敏 张丽 张兴 袁龙 杨世军

摘 要：为解决城市交通中存在的停车难问题，本研究提出基于计算机视觉和机器学习软件库（OpenCV）的地面停车诱导系统，以某露天停车场为例，开展实地测试，分析设备的拍摄角度、拍摄高度和级联分类器个数对设备识别效果的影响。试验结果表明，设备的拍摄角度为0°、拍摄高度为10 m、级联分类器个数为6个，设备的识别效果最为理想;设备对除黑色车型的识别率（0.77）较低外，对其他颜色车型的识别率均在0.85以上，其中对白色车辆的识别率最高（0.92），车辆颜色与停车场地面颜色色差越大时，设备识别率越高。

关键词：交通工程;停车诱导系统;计算机视觉;车辆检测;移动通信

中图分类号：U491    文献标识码：A   文章编号：1006-8023（2021）03-0119-07

Abstract：In order to solve the parking difficult problems of urban traffic， this study proposed a ground parking guidance system based on computer vision and machine learning library （OpenCV）. Taking an open parking lot as an example， field tests were carried out to analyze the effects of the shooting angle， shoot height and the number of cascade classifiers on the recognition effect of equipment. The test results showed that the best recognition effect was achieved when the shooting angle of the device was 0 degrees， the shooting height was 10m and the number of cascading classifiers was 6. Except for the low recognition rate （0.77） of black car， the recognition rate of other color cars was above 0.85， and the recognition rate for white car was the highest （0.92）. The greater the color difference between the vehicle color and the parking lot ground color， the higher the device recognition rate.

Keywords：Traffic engineering; parking guidance system; computer vision; vehicle detection; mobile communication

0 引言

國民经济水平提高使得汽车保有量随之提高，但停车场的建设速度与汽车保有量的增长速度无法匹配，由此导致“停车难”等问题[1]。“车多车位少”的现象愈演愈烈的同时，车位难寻的问题也不容忽视。驾驶人无法准确获知停车场内车位利用情况，难以快速地找到空闲车位，车辆在停车场内盲目行驶导致拥堵，以及排队缴费浪费时间等[2]。这造成了路外公共停车场利用率低下，驾驶人更倾向于将车辆停放于路内停车场。车辆停放过程中对动态交通产生干扰的同时，易出现车辆追尾、剐蹭等安全事故[3-4]。停车诱导目前存在诸如安装于路侧的LED停车诱导屏安装难度较大，建设成本偏高;同时驾驶员在注视诱导屏的过程中，视觉分心会产生安全隐患[5]。……