基于场景感知的运动目标检测方法

发布时间：2018-05-19 01:29

本文选题：目标检测 + 场景感知　；参考：《电子学报》2016年11期

【摘要】：背景减除法是一种主要的运动目标检测框架,但在复杂环境中构建一种初始模型建立周期短、可靠性高、鲁棒性好的模型仍是一大难题.本文从场景感知的角度出发,在背景减除框架的基础上提出一种目标检测方法.该方法根据前两帧中稳定的结构信息感知背景中潜在的前景区域,在第二帧建立初始模型时利用最近邻域背景像素点代替可能的前景像素点,提高了初始模型可靠性;结合颜色信息和二进制特征提出了像素点二级分类判决机制,并通过感知像素点邻域内的纹理复杂度自适应调整局部判决阈值和更新频率;在模型更新阶段提出处理误判的反馈机制.在公开视频序列上同几种流行检测算法的实验对比结果证明了本文算法的有效性和优越性.
[Abstract]:Background subtraction is a main framework for moving target detection, but it is still a difficult problem to construct an initial model with short period, high reliability and good robustness in complex environment. In this paper, a target detection method based on background subtraction framework is proposed from the perspective of scene perception. According to the stable structure information in the first two frames, the method perceives the potential foreground region in the background, and uses the nearest neighborhood background pixels instead of the possible foreground pixels in the second frame to improve the reliability of the initial model. Combining the color information with binary features, a two-level pixel classification and decision mechanism is proposed, and the local decision threshold and update frequency are adjusted adaptively by sensing the texture complexity in the pixel neighborhood. At the stage of model updating, a feedback mechanism to deal with misjudgment is proposed. The experimental results of several popular detection algorithms on open video sequences show that the proposed algorithm is effective and superior.
【作者单位】：信息工程大学信息系统工程学院;南京理工大学自动化学院;
【基金】：国家科技重大专项(No.2014ZX03006003)
【分类号】：TP391.41

【相似文献】