基于最小特征匹配代价的夜间车辆检测和追踪  

NIGHTTIME VEHICLE DETECTION AND TRACKING BASED ON MINIMUM FEATURE MATCHING COST

作　　者：汤春明[1] 聂美玲 丁腾飞[2] 浩欢飞 韩旭[2] 

机构地区：[1]天津工业大学电子与信息工程学院,天津300387 [2]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《计算机应用与软件》2015年第4期292-296,共5页Computer Applications and Software

摘　　要：针对夜间复杂环境下车辆追踪存在检测难、配对准确率低、追踪效果差等问题,提出一种新的夜间车辆检测与追踪算法。首先将频域的同态滤波与空域的阈值技术结合进行车灯检测;然后,利用几何特征对车灯进行跟踪;其次,将几何特征和运动特征相结合,利用最小特征匹配代价算法实现车灯配对;最后,根据车灯配对情况对车辆轨迹进行追踪,同时引入反馈修正机制对轨迹进行修正。实验表明该算法能够在不同照明和交通条件下有效检测车灯、跟踪车辆,平均检测率和跟踪率较高。Given there are many problems in vehicle tracking under complex environment at nighttime such as difficult in detecting,low accuracy of matching and poor tracking effect,in the paper we propose a new nighttime vehicle detection algorithm.First,the homomorphism filter in frequency domain is combined with the threshold technique in spatial domain for vehicle lights detection,and the geometric feature is used to track the headlights.Secondly,the algorithm combines the geometric features with motion features,and employs minimum feature matching cost algorithm to achieve headlights matching.Finally,according to headlights matching situation it tracks the traces of vehicles and corrects the traces by introducing the feedback correcting mechanism.Experiments show that the proposed method can detect headlights and track vehicles effectively in different lighting and traffic conditions,the average detection rate and tracking rate are relatively high.

关 键 词：车辆监测 车辆追踪 最小特征匹配代价 反馈机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]