交通灯自动检测技术研究

发布时间：2018-06-13 04:27

本文选题：交通灯检测 + 形态学处理　；参考：《武汉理工大学》2014年硕士论文

【摘要】：随着经济高速发展，汽车也越来越普及。然而，汽车在带来便利的同时，也使得城市交通状况更加复杂，这就需要更多的交通灯为交通路口提供导航信息。而智能车在行驶过程中同样需要获取交通灯信息来进行相应的行为决策。在此，本文提出了一种对交通灯进行检测分类的算法。和主流的检测算法所不同的是，本系统先对交通灯背板进行检测，然后再根据发光单元的相关特性进行检测与分类。为了提高运行速度，该算法首先对天空等色度与背板差异较大的区域进行筛除；然后利用交通灯背板的相关特性进行筛选，得到交通灯背板目标区域；最后利用交通灯发光单元的特性进行定位，并识别出发光单元的类型。本文的主要工作和贡献如下。（1）对于经过预处理后的图像，，合理的利用数学形态学操作对交通灯背板和与之相连的横杆进行分离，并根据背板的基本几何特征对候选区域进行过滤。（2）采用基于主分量分析的最大似然比检验算法区分交通灯背板和杂质，以提供合理的背板目标区域供后续发光单元检测算法进行定位分类。（3）在对交通灯发光单元的类型进行区分时，将Hu不变矩、Hough圆检测和圆形度检测进行对比分析，并选出合适的算法提取圆形交通灯；对于箭头形交通灯，在比对了模板匹配和坐标轴投影检测方法的实验效果之后选择后者进行快速高效的分割。对实景车辆采集到的图片进行测试，结果验证了本系统在红灯检测中的可行性和高效性。对由于图像质量等其它不确定因素导致的误检和漏检，本论文在最后也给出了未来工作的改进思路。
[Abstract]:With the rapid development of economy, cars are becoming more and more popular. However, while the automobile brings convenience, it also makes the urban traffic situation more complicated, which requires more traffic lights to provide navigation information for traffic junctions. The intelligent vehicle also needs to obtain the traffic light information to make the corresponding behavior decision. In this paper, a traffic light detection and classification algorithm is proposed. Different from the mainstream detection algorithm, the system detects the backplane of the traffic light first, then detects and classifies the backplane according to the characteristics of the light-emitting unit. In order to improve the running speed, the algorithm firstly sieves out the regions with big difference between the sky chroma and the backplane, and then sift through the related characteristics of the traffic light backplane to obtain the target area of the traffic light backplane. Finally, the characteristics of the traffic light emitting unit are used to locate and identify the type of the light emitting unit. The main work and contribution of this paper are as follows. The candidate regions are filtered according to the basic geometric features of the backplane. (2) the maximum likelihood ratio test algorithm based on principal component analysis (PCA) is used to distinguish the traffic light backplane from the impurity. In order to provide a reasonable backplane target area for the subsequent luminous unit detection algorithm to locate and classify. 3) when the traffic light emitting unit type is distinguished, the Hu invariant moment Hough circle detection and the circular degree detection are compared and analyzed. An appropriate algorithm is selected to extract circular traffic lights. For arrowhead traffic lights, the experimental results of template matching and coordinate axis projection detection are compared and the latter is selected for fast and efficient segmentation. The results show that the system is feasible and efficient in red light detection. At the end of this paper, the improvement ideas of future work are also given for the false detection and missed detection caused by other uncertain factors such as image quality.
【学位授予单位】：武汉理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;TP391.41

【参考文献】