摘要
近年来,光流法被广泛应用到运动目标检测中,但该类方法计算量大,不利于实时处理,且易受噪声影响.提出一种结合运动边界和稀疏光流的运动目标检测方法.该方法通过结构随机森林提取图像的运动边界,然后计算运动边界的光流矢量从而实现目标检测.由于运动边界不易受噪声影响且其光流矢量计算满足稀疏性,将运动边界和光流法结合,保证了算法的鲁棒性和实时性.实验结果表明,该方法可以精确地检测出完整的运动目标,且具有较低的计算复杂度,有利于实时性运动目标检测系统的构建.
Optical flow approaches have been widely used in moving object detection in recent years. However, most of these methods suffer from a heavy calculation loads and poor real-time performance. To solve these issues ,a novel moving object detection technique based on motion boundaries and sparse optical flow is presented. The proposed method employs the structured random forest to extract motion boundaries and detects the target objects by calculating the optical flow vectors of motion boundaries. Since the motion bounda- ries are robust to noise and optical flow Vectors are sparse,the proposed method possesses the better robustness and low computation cost. Experimental results show that the presented approach is capable of detecting moving objects accurately and completely with lower time cost, which is beneficial for real-time detection.
出处
《小型微型计算机系统》
CSCD
北大核心
2017年第3期635-639,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61201435
61473118)资助
湖南省教育厅开放基金项目(15K051)资助
湖南省高校科技创新团队支持计划项目(湘教通[2012]318号)资助
关键词
运动目标检测
运动边界
稀疏光流
结构随机森林
moving object detection
motion boundaries
sparse optical flow
structured random forest
