智能汽车交通灯识别方法研究

发布时间：2018-04-04 20:47

本文选题：交通信号灯　切入点：机动车制动灯　出处：《哈尔滨工业大学》2016年硕士论文

【摘要】：智能汽车通过辅助或代替驾驶员操作,以提高乘车舒适性和安全性,是当前汽车行业的研究热点。但无人驾驶技术尚未成熟,且交通基础设施升级周期较长,因此现阶段智能汽车主要通过搭载辅助驾驶系统实现智能化。交通灯识别技术作为辅助驾驶系统技术的一部分,是智能汽车在城市道路上安全行驶的关键技术之一。本文针对交通信号灯以及机动车制动灯的识别方法进行研究,针对日间和夜间工况的特点,分别设计了识别方法,并使用Matlab平台进行验证。首先,针对日间圆形交通信号灯的识别,设计了识别方法。对原图像进行基于RGB空间的阈值筛选,在颜色筛选结果的基础上,通过基于形态特征的区域形态特征筛选,实现对圆形交通信号灯的识别。针对日间箭头形交通信号灯的识别,设计了基于RGB空间的阈值筛选和基于局部灰度特征的箭头形区域提取方法,提取出完整箭头形区域,在候选区域提取结果的基础上,通过区域形态特征以及重心偏移特征,识别候选区域中的箭头形交通信号灯。其次,针对夜间交通信号灯的特殊工况设计了识别方法。对原图像进行基于RGB空间的阈值筛选,通过区域包围零点数量的筛选以及局部区域取反操作,消除颜色阈值筛选结果中存在的不规则光晕,提取出完整的交通信号灯轮廓,通过基于形态特征的区域形态特征筛选方法,结合Canny边缘检测和Hough变换检测其中的圆形轮廓,完成进一步的识别。最后,针对日间及夜间的机动车制动灯识别,提出了相应的识别方法。日间机动车制动灯识别,主要基于RGB空间的阈值筛选,以及基于图像平均灰度的阈值筛选。对颜色灰度双阈值筛选的结果,通过形态学膨胀运算增强。夜间机动车制动灯识别,对原图像采用YCbCr空间的阈值筛选,并对原图像进行Top-Hat变换结合最大类间方差法的处理,提取图像中的光源区域。通过对颜色筛选结果和光源区域提取结果进行形态学处理及融合,提取出制动灯区域。对日间及夜间的制动灯识别结果均进行配对处理,标记同一辆机动车的制动灯,完成制动灯的识别过程。对交通信号灯和机动车制动灯识别方法进行实验,结果表明,本文所提出的方法能够有效识别出图像中的交通灯,但是对某些复杂情况则无法完全正确识别,仍存在着改进空间。
[Abstract]:Intelligent vehicle is the research hotspot of automobile industry at present, which can improve ride comfort and safety by auxiliary or replacing driver operation.However, the driverless technology is not mature, and the transportation infrastructure upgrade cycle is long, so the intelligent vehicle is mainly realized by using the auxiliary driving system at the present stage.As a part of auxiliary driving system, traffic light recognition is one of the key technologies for intelligent vehicles to drive safely on urban roads.In this paper, the identification methods of traffic lights and vehicle brake lights are studied. According to the characteristics of daytime and night conditions, the identification methods are designed and verified by using Matlab platform.Firstly, a recognition method is designed for the recognition of daytime circular traffic lights.On the basis of the color filtering results, the recognition of circular traffic lights is realized by the morphological feature based region morphological feature selection of the original image.Aiming at the recognition of arrowhead traffic lights in daytime, a threshold screening method based on RGB space and an arrowhead region extraction method based on local gray level feature are designed to extract the complete arrowhead region, and based on the result of candidate region extraction.The arrowhead traffic lights in candidate regions are identified by region shape feature and barycenter shift feature.Secondly, the identification method is designed for the special working condition of night traffic light.The original image is screened based on the threshold value in RGB space. The irregular halo in the color threshold screening result is eliminated by the selection of the number of surrounding zeros in the region and the inverse operation of the local area, and the complete contour of the traffic signal is extracted.By means of morphological feature based region morphological feature selection method, combined with Canny edge detection and Hough transform to detect the circular contour, the further recognition is completed.Finally, a corresponding recognition method is put forward for vehicle brake lamp recognition in daytime and night.The recognition of daytime vehicle brake lamp is mainly based on the threshold screening in RGB space and the threshold screening based on the average gray scale of the image.The results of color grayscale double threshold screening were enhanced by morphological expansion operation.In the recognition of vehicle brake lamp at night, the threshold value of YCbCr space is used to filter the original image, and the original image is processed by Top-Hat transform combined with the method of maximum inter-class variance to extract the region of light source in the image.The area of brake lamp was extracted by morphological processing and fusion of color screening results and light source region extraction results.The identification results of day and night braking lights were matched to mark the brake lights of the same vehicle to complete the identification process of the brake lights.The results of experiments on traffic signal lights and vehicle brake lights show that the method proposed in this paper can effectively recognize traffic lights in the image, but it can not be fully recognized for some complicated cases.There is still room for improvement.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6;TP391.41

【相似文献】