CAJ | 학술논문

运动目标检测是计算机视觉、模式识别、视频编码、基于内容的检索等领域的研究重点和难点，在视频监控、虚拟现实、机器人导航等许多领域得到广泛应用。针对运动目标检测中的动态背景问题，提出一种多分辨率互补检测方法。首先，通过对图像进行降低分辨率和高斯模糊来减少动态纹理和背景扰动的影响，在低分辨率下对图像进行区域分割，基于直方图特征使用IIR滤波器法建立低分辨背景模型，并检测出可能存在运动目标的区域；然后，对于高分辨率图像，基于自适应的混合高斯模型，对低分辨率下检测到的可能存在运动目标的对应区域进行目标检测，修正低分辨率的背景模型，实现多分辨率互补检测过程。
운동목표검측시계산궤시각、모식식별、시빈편마、기우내용적검색등영역적연구중점화난점，재시빈감공、허의현실、궤기인도항등허다영역득도엄범응용。침대운동목표검측중적동태배경문제，제출일충다분변솔호보검측방법。수선，통과대도상진행강저분변솔화고사모호래감소동태문리화배경우동적영향，재저분변솔하대도상진행구역분할，기우직방도특정사용IIR려파기법건립저분변배경모형，병검측출가능존재운동목표적구역；연후，대우고분변솔도상，기우자괄응적혼합고사모형，대저분변솔하검측도적가능존재운동목표적대응구역진행목표검측，수정저분변솔적배경모형，실현다분변솔호보검측과정。
Nowadays motion detection under dynamic background has been an important and difficult research area of computer vision and video processing. For motion detection in the dynamic context, a complementary multi-resolution detection method is proposed. Dynamic textures and background disturbances in images are reduced by lowering the image resolution and applying Gaussian blur. First, the low-resolution image is segmented into blocks, and then the background model is built using histogram characteristics obtained by IIR filters to detect the possible areas that moving target probably exists;then, the motion detection is executed on the high-resolution image based on adaptive Gaussian model in the corresponding detected regions in the low resolution image, and the learning rates of low-resolution model are refined with the detection results, so the complementary multi-resolution detection process is implemented.