交通信号灯检测、跟踪、定位与识别方法研究

发布时间：2018-04-15 00:10

本文选题：信号灯识别 + 目标跟踪　；参考：《北京理工大学》2015年硕士论文

【摘要】：交通信号灯识别是无人驾驶技术的一个基本组成部分。本课题的研究内容是基于无人驾驶车辆平台上丰富的传感器与信息资源，系统地设计一套包括了检测、跟踪、定位、识别功能的交通信号灯识别系统。本课题设计的信号灯识别系统所融合的传感器及信息包括：车载相机所采集的图像序列，全球定位系统/惯导组合导航系统（即GPS/INS组合导航系统）所测量的，或者通过激光雷达和即时定位与地图构建算法（即SLAM算法）所获得的车辆平台位置与航向。本课题中，跟踪模块将处理图像序列的检测器的输出看作是交通信号灯目标的受干扰带噪声观测，而且考虑到检测所得的候选区域可能出现误检与漏检的情况。跟踪器利用多目标跟踪算法中的数据关联对这些观测进行处理，从杂乱的观测中提取来自信号灯的真实候选区域并剔除误检的区域。经过数据关联后的稳定候选区域将会经过分类器，识别信号灯的图案。在这个总的流程之下，为了提高多目标跟踪的性能，本课题对信号灯这一特殊目标的运动模型做了比较深入的研究，设计了运动预测算法。在车辆平台接近路口时，信号灯的三维位置对运动模型有较大影响，所以课题又设计了基于多视角观测的交通信号灯定位算法。而跟踪与定位的准确性又与车辆自身的位置、姿态测量相关，为了弥补GPS/INS组合导航系统的缺陷，，课题又对非外源式的SLAM算法进行了研究，提出来一种基于图像处理的SLAM算法。此外，由于该系统融合了来自多个不同传感器的多种数据，所以也对传感器间的时间、空间对准问题作了研究。
[Abstract]:Traffic signal recognition is a basic component of driverless technology.Based on the abundant sensors and information resources on the driverless vehicle platform, this paper systematically designs a set of traffic signal light recognition system which includes detection, tracking, positioning and recognition functions.The sensors and information of the signal lamp recognition system designed in this paper include: the image sequence collected by the vehicle camera, the global positioning system (GPS) / inertial navigation integrated navigation system (GPS/INS integrated navigation system),Or the vehicle platform position and course can be obtained by lidar and real-time location and map construction algorithm (i.e. SLAM algorithm).In this thesis, the tracking module regards the output of the detector which processes image sequence as the noise observation of the target of traffic signal light, and considers that the candidate region of the detection may appear the case of false detection and miss detection.The tracker uses the data association in the multi-target tracking algorithm to process these observations, and extracts the true candidate regions from the signal lights from the clutter observations and removes the areas of false detection.After data association, the stable candidate area will be identified by classifier to recognize the pattern of the signal light.In this general process, in order to improve the performance of multi-target tracking, the motion model of signal lamp is studied deeply, and the motion prediction algorithm is designed.When the vehicle platform is near the intersection, the three-dimensional position of the signal light has great influence on the motion model, so the algorithm of the traffic signal lamp location based on multi-angle observation is designed.The accuracy of tracking and positioning is related to the position and attitude measurement of the vehicle itself. In order to make up for the defects of GPS/INS integrated navigation system, the non-exogenous SLAM algorithm is studied, and a SLAM algorithm based on image processing is proposed.In addition, the time and space alignment between sensors is also studied because the system integrates many kinds of data from many different sensors.
【学位授予单位】：北京理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41

【参考文献】