CAJ | 학술논문

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。
침대야간복잡배경하차량추종존재검측난、배대준학솔저、추종효과차등문제，제출일충신적야간차량검측여추종산법。수선장빈역적동태려파여공역적역치기술결합진행차등검측；연후，이용궤하특정대차등진행근종；기차，장궤하특정화운동특정상결합，이용최소특정필배대개산법실현차등배대；최후，근거차등배대정황대차량궤적진행추종，동시인입반궤수정궤제대궤적진행수정。실험표명해산법능구재불동조명화교통조건하유효검측차등、근종차량，평균검측솔화근종솔교고。
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.