基于改进ViBe的自适应运动目标检测算法
Adaptive moving object detection algorithm based on improved ViBe
费莉梅 1田翔 2郑博仑3
作者信息
- 1. 之江实验室智能装备研究院,浙江杭州 311100
- 2. 浙江大学生物医学工程与仪器学院,浙江杭州 310027
- 3. 杭州电子科技大学自动化学院,浙江杭州 310018
- 折叠
摘要
针对ViBe算法无法去除动态背景,易出现鬼影及不能自适应光照变化的问题,提出一种复杂环境自适应的ViBe改进算法.通过计算区域的复杂度、闪烁波动度,对分类半径R和更新率T进行动态调整,对样本点进行有效性权重的计算,更高效地过滤背景噪声和适应光照渐变;在检测物体状态变化时,动态调整R和T,通过融合前景点计数和帧差法优化鬼影消除;通过识别最小外接矩阵区域差异加快去除鬼影;利用帧差法实时检测光照突变,及时进行重新初始化,避免大量误检.实验结果表明,改进ViBe算法在适应动态背景、光照变化及抑制鬼影等方面比原算法均有更好检测效果,检测精度平均提升了 40.7％.
Abstract
Aiming at the problems that the ViBe algorithm is prone to ghosting,cannot remove dynamic backgrounds and cannot adapt to sudden changes in illumination,an improved ViBe algorithm that was adaptive to complex environments was proposed.The classification radius R and update rate T were dynamically adjusted by calculating the complexity and flickering fluctuation of the area,and the validity weight of the sample points was calculated at the same time to filter background noise and accommodate lighting gradients effectively.When the moving state of the detected running object changed suddenly,the R and T were dynami-cally adjusted,and the ghost elimination was optimized by integrating the foreground point count and frame difference method.The ghost area was judged by comparing the minimum circumscribed matrix location difference identified using different methods to further speed up ghost removal.The frame difference method was used to detect the sudden changing in the illumination in real time,and to re-initialize the model in time to avoid a large number of false detections.Experimental results show that the improved ViBe algorithm has better detection results than the original algorithm in terms of removing ghosts,adapting to chan-ges in illumination and complex dynamic backgrounds,and the detection accuracy is improved by 40.7％on average.
关键词
ViBe算法/运动目标检测/复杂背景/自适应阈值/动态场景/鬼影消除/背景建模/自适应Key words
ViBe algorithm/moving object detection/complex background/self-adaptive threshold/dynamic scene/ghost elimi-nation/background modeling/adaptive引用本文复制引用
出版年
2024
NETL
NSTL
万方数据