基于改进混合高斯模型与阴影去除的目标检测

Object Detection Based on the improved Gaussian mixture model and shadow removal

摘要:

随着计算机视觉和摄像设备的日益普及,目标检测技术已经成为一个重要的研究领域。虽然提出了几种目标检测方法,但由于其适用性与局限性,并不能解决实际复杂场景中的各种挑战。针对传统混合高斯模型对动态背景、光照变化和阴影敏感等问题,提出一种混合高斯模型的改进算法,用于视频中目标检测。该方法首先通过传统混合高斯模型获取当前帧目标的粗略区域；通过将双级学习率和组合权重引入混合高斯模型,从而区分出运动区域与包含动态背景的背景区域；然后进一步利用基于颜色特性与空间连续性的方法去除阴影；最后通过形态学处理提取出准确的运动目标区域。对比实验表明,所提方法不仅能够有效去除动态背景,而且能够有效抑制阴影和光照变化的影响。

Abstract:

Object detection technologies have emerged as an important research area with increasing popularity of computer vision and camera devices. Even though several object detection approaches have been proposed, they cannot address various challenges in actual complex scenes owing to their applicability and restrictions. For the traditional Gaussian mixture model is sensitive to dynamic background, light change and shadow, an improved algorithm of Gaussian mixture model is proposed, which is used for object detection in video. Firstly, we extract rough region of the current frame by the traditional Gaussian mixed model. By introducing the two-level learning rate and combined weight into the Gaussian mixture model, the moving region and the background region containing the dynamic background are distinguished. Then shadow is removed based on color characteristic and spatial continuity. Finally, the complete and accurate moving object area is detected out by morphological closing operation. Comparative experiments indicate that the proposed method not only can effectively restrain the influence of dynamic background, but also suppress the influence of shadow and light changes.