基于车载视觉系统的目标检测优化算法研究
发布时间：2019-03-31 12:05
【摘要】：随着我国社会经济的迅速发展,我国汽车保有量大大增加,但与此同时交通事故的发生率也逐渐上升。高级驾驶辅助系统(ADAS)是解决交通安全问题的重要手段之一,成为研究学者们所关注的重要研究课题。目标检测算法作为高级驾驶辅助系统中的关键技术之一,近年来有价值的研究成果层出不穷,Subcat,RCNN,Faster-RCNN,YOLO等目标检测算法在简单场景下有着不错的表现,但是若将这些检测算法应用于实际的交通场景仍存在一定的局限性。本文对复杂交通场景下驾驶辅助系统的实际应用问题进行了研究,提出了提高现有目标检测算法精度的优化方案。论文的主要工作如下:1.分析了现有目标检测算法在复杂交通场景中误检增多问题的成因,本文根据摄像机成像原理,提出了利用几何约束模型去除误检的方法。2.针对现有目标检测算法中出现的漏检和检测位置不准确问题,本文提出了基于条件随机场(CRF)的连续运动信息融合模型,进而提高了目标检测算法的性能。3.通过对比实验验证了本文提出的目标检测优化算法的有效性。实验结果显示在不同的复杂路况场景下,本文提出的目标检测优化算法仍具有可靠性。本文提出的优化算法可以与现有的多种目标检测算法相结合,为复杂交通场景下的目标检测问题提供了新思路,推进了自动驾驶领域的研发进程。
[Abstract]:With the rapid development of China's social economy, the number of cars in our country increases greatly, but at the same time, the incidence of traffic accidents is also increasing gradually. Advanced driving Assistance system (ADAS) is one of the important means to solve the traffic safety problem, which has become an important research topic concerned by scholars. Target detection algorithm is one of the key technologies in the advanced driving assistant system. In recent years, valuable research results emerge one after another. Target detection algorithms such as Subcat,RCNN,Faster-RCNN,YOLO have a good performance in simple scenes. However, if these detection algorithms are applied to the actual traffic scene, there are still some limitations. In this paper, the practical application of the driving assistant system in complex traffic scene is studied, and the optimization scheme to improve the accuracy of the existing target detection algorithms is proposed. The main work of this paper is as follows: 1. Based on the principle of camera imaging, this paper puts forward a method to remove false detection by geometric constraint model. 2.2.Based on the principle of camera imaging, this paper proposes a method to remove false detection by geometric constraint model. In this paper, a continuous motion information fusion model based on conditional random field (CRF) is proposed to improve the performance of target detection algorithms. 3. The effectiveness of the proposed optimization algorithm for target detection is verified by comparison experiments. The experimental results show that the target detection optimization algorithm proposed in this paper is still reliable under different complex road conditions. The optimization algorithm proposed in this paper can be combined with a variety of existing target detection algorithms, which provides a new idea for target detection in complex traffic scenarios and advances the research and development process in the field of autopilot.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41
本文编号：2450885
[Abstract]:With the rapid development of China's social economy, the number of cars in our country increases greatly, but at the same time, the incidence of traffic accidents is also increasing gradually. Advanced driving Assistance system (ADAS) is one of the important means to solve the traffic safety problem, which has become an important research topic concerned by scholars. Target detection algorithm is one of the key technologies in the advanced driving assistant system. In recent years, valuable research results emerge one after another. Target detection algorithms such as Subcat,RCNN,Faster-RCNN,YOLO have a good performance in simple scenes. However, if these detection algorithms are applied to the actual traffic scene, there are still some limitations. In this paper, the practical application of the driving assistant system in complex traffic scene is studied, and the optimization scheme to improve the accuracy of the existing target detection algorithms is proposed. The main work of this paper is as follows: 1. Based on the principle of camera imaging, this paper puts forward a method to remove false detection by geometric constraint model. 2.2.Based on the principle of camera imaging, this paper proposes a method to remove false detection by geometric constraint model. In this paper, a continuous motion information fusion model based on conditional random field (CRF) is proposed to improve the performance of target detection algorithms. 3. The effectiveness of the proposed optimization algorithm for target detection is verified by comparison experiments. The experimental results show that the target detection optimization algorithm proposed in this paper is still reliable under different complex road conditions. The optimization algorithm proposed in this paper can be combined with a variety of existing target detection algorithms, which provides a new idea for target detection in complex traffic scenarios and advances the research and development process in the field of autopilot.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41
【参考文献】
相关期刊论文 前9条
1 本刊讯;;百度携手宝马开发无人驾驶车[J];中国公共安全;2014年19期
2 杨帆;;无人驾驶汽车的发展现状和展望[J];上海汽车;2014年03期
3 马钧;曹静;;基于中国市场特定需求的汽车先进驾驶辅助系统发展趋势研究[J];上海汽车;2012年04期
4 王牛;李祖枢;武德臣;于芳;;机器人单目视觉定位模型及其参数辨识[J];华中科技大学学报(自然科学版);2008年S1期
5 李政;刘大学;贺汉根;;触发式车辆实时检测新算法[J];华中科技大学学报(自然科学版);2008年S1期
6 胡铟;杨静宇;;基于模型的车辆检测与跟踪[J];中国图象图形学报;2008年03期
7 胡文;孙容磊;;基于序列图像的视觉定位与运动估计[J];传感器与微系统;2007年07期
8 王荣本,游峰,崔高健,郭烈;基于计算机视觉高速智能车辆的道路识别[J];计算机工程与应用;2004年26期
9 徐友春,王荣本,李兵,李斌;世界智能车辆近况综述[J];汽车工程;2001年05期
相关硕士学位论文 前1条
1 衡浩;复杂交通场景中基于路面提取的行人检测研究[D];上海交通大学;2013年
，本文编号：2450885
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2450885.html
