基于协同训练的低空运动平台动态人物阴影检测

Detecting Dynamic Human Shadow Based on Low Altitude Moving Platform by Co-training

针对现有视频监控中人物阴影检测大都采用背景减法,难以实现低空运动平台下的动态检测问题,提出一种针对低空运动平台的动态人物阴影检测方法.在改进现有的3种室外人物阴影像素特征的基础上,提出一种亮度反差区域特征,并通过实验给出了其优化组合模式;基于像素与区域特性的独立性构建双视图分类器,设计了与之相适应的半监督协同训练策略;最后针对实时处理需求,提出了通过随机采样改善学习效率、利用支持向量机解决小样本学习问题的加速方案.实际低空运动平台下的实验结果表明,该方法具有较高的人物阴影检测率与较好的算法鲁棒性,可有效地解决低空运动平台下的高质量动态人物阴影检测问题.

In the area of video-monitoring,background subtraction method is widely adopted to detect human shadows,but it is difficult to achieve dynamic detection for the low altitude moving platforms.In order to resolve the problem of dynamic human shadow detection onboard of low altitude moving platform,a novel method is proposed in this paper.In this method,according to the characteristics of the outdoor human shadow,three kinds of pixel based features are improved,a kind of area based feature named ＂luminance contrast＂ is proposed,and further more,an optimized combination of those features is given experimentally.Secondly,according to the independence between pixel based and area based features,a two-view classifier based on co-training theory and a semi-supervised training strategy for this classifier are established.And then,the random sampling theory is adopted for improving the training efficiency and the support vector machine has also been adopted to solve the problem of small samples learning.Some experimental results show that this proposed method is characterized by high shadow detection rate and fine robustness,can effectively solve the problem of dynamic human shadow detection for low altitude moving platforms.