CAJ | 학술논문

根据环境照度将夜间交通场景区分为充足照明和低照度两种情况，分别设计相应的处理流程检测运动车辆。针对充足照明的情况，先使用梯度滤波消除路面反光的干扰，再进行帧间差分检测运动区域，最后使用级联形态学滤波消除噪声点和填充帧间差分方法导致的车体区域空洞。针对低照度情况，引入先验知识检测车前灯，并利用车灯间距判别车型大小，最后结合车辆的造型知识定位车体。对多段典型的夜间交通场景视频进行了测试，实验结果表明，该方法能够有效地检测夜间行驶车辆。
근거배경조도장야간교통장경구분위충족조명화저조도량충정황，분별설계상응적처리류정검측운동차량。침대충족조명적정황，선사용제도려파소제로면반광적간우，재진행정간차분검측운동구역，최후사용급련형태학려파소제조성점화전충정간차분방법도치적차체구역공동。침대저조도정황，인입선험지식검측차전등，병이용차등간거판별차형대소，최후결합차량적조형지식정위차체。대다단전형적야간교통장경시빈진행료측시，실험결과표명，해방법능구유효지검측야간행사차량。
Night traffic scenes are distinguished as two cases of good lighting and poor visibility according to the environment illumination. The corresponding strategies are schemed to deal with vehicle detection for each case. In the former case, a preprocessing based on grad-filtering is firstly employed to eliminate the influence of the reflection. Foregrounds are then detected by inter-frame differencing. Finally, a post-processing based on a cascaded morphological filter is exploited to exclude the noises and fill the holes resulting from the operator of inter-fi＇ame differencing. In the other case, a priori knowledge is introduced to detect vehicle headlights pairs, and vehicle body is located based on the knowledge of vehicle configuration. Experiments are done on several video sequences representing typical night traffic scene both in the two cases. Results show that the presented methodology is able to detect vehicles at night effectively.