摘要
视频动态目标检测与跟踪是智能化视频分析的基础,是实现智能监控的关键技术之一。基于人类视觉对运动的方向和速度非精确感知的特点,结合HR生物相关运动检测模型改进Itti Saliency算法,建立颜色、方向、亮度和运动四特征通道的特征图提取算法,对特征图进行跨尺度融合及归一化,从而提取视频图像中动态目标的视觉显著图。对视频序列图像的显著图逐一显示,便可实现对运动目标的跟踪。提出的运动感知模型,改善了对运动目标视觉显著性的检测效果,能够准确检测并跟踪监控视频中复杂背景、遮挡、多物体的动态目标。
Video dynamic target detection and tracking is the basis for intelligent video analysis and is the key technology to achieve intelligent monitoring. Based on the characteristic of non-precision sensing of the human visual for direction and speed of the movement, Itti Saliency algorithm is improved by combined with HR algorithm, which is a bio-related motion detection mode. The characteristic map extraction algorithm is presented by creating color, direction, brightness and motion characteristic channels. Then the visual saliency map of dynamic targets in video images is generated by fusing and normalizing of different scale characteristic maps. At last, the dynamic targets are detected and tracked by displaying saliency maps one by one in video image sequence. The experimental results show that proposed motion perception model improves the detection effect of visual saliency for dynamic targets and is able to be applied in clutter background, occlusion and multiple target tracking problems.
出处
《信息技术》
2014年第4期60-65,共6页
Information Technology
基金
重庆市基础与前沿研究计划项目(cstc2013jcyjA40038)
重庆市公安局科技攻关项目(2012-11)
重庆警察学院自然科学研究项目(jy20128004)