融合时空特征的自适应运动目标检测方法

Adaptive moving object detection method based on spatiotemporal features

视频序列中的运动目标检测通常是作为更高级别图像分析与应用的基础。大多数目标检测算法是基于背景减法的,通过对含有历史背景像素值的背景模型进行分析来完成检测,但在动态场景中存在误检的问题,如风摇动树和流动河流。为了提高运动目标检测的精确性和降低误检率,提出了一种融合时空特征的自适应运动目标检测方法。该方法融合颜色与纹理特征进行时空建模,通过分析视频序列来检测动态背景的复杂度及动态背景区域,基于像素级反馈环来自适应调整模型的内部参数,替代手动设置的全局参数,提高模型的灵敏度和场景适应速度。在多个视频序列上进行对比,测试结果表明本文方法精确率高、错误分类比低,对较复杂场景具有一定的鲁棒性。

Moving object detection in video sequences is usually used as the basis for higher-level image analysis and applications. Most object detection algorithms are based on background subtraction. The detection is performed by analyzing the background model containing historical background pixel values. However, there are problems of false detection in dynamic scenes, such as wind shaking trees and flowing rivers. In order to improve the accuracy of moving objec detection and reduce the false detection rate, proposes an adaptive moving object detection method based on spatiotemporal features. This method combines color and texture features for spatiotemporal modeling. It analyzes the video sequence to detect the complexity of the dynamic background and the dynamic background area. The pixel-level feedback loop is used to adaptively adjust the internal parameters of the model, instead of manually setting global parameters, and improving the model′s sensitivity and scene adaptation speed. The comparison test results on multiple video sequences show that our method has high accuracy, low misclassification ratio, and is robust to more complex scenes.