基于自适应混合高斯模型全方位视觉目标检测

针对全方位视觉传感器视野范围大的特点,提出一种基于自适应混合高斯模型的全方位视觉目标检测系统.该系统通过Hough变换检测全方位图像的中心,基于图像中心对全方位图像进行展开.对展开后的图像利用混合高斯模型进行背景建模,并自适应地更新背景模型,通过前景分割可以有效地分割出运动目标.在图像展开及混合高斯建模时,通过调整系统的采样频率可以较好地改善目标检测的实时性.实验结果表明,该系统可以在复杂环境中有效地检测运动目标,具有较强的准确性和鲁棒性.

Adaptive mixture Gaussian model based target detection using omnidirectional vision

Considering the advantage of omnidirectional vision sensor with large field of view, an omnidirectional vision target detection system was presented based on the adaptive mixture Gaussian model. A Hough transform algorithm was adopted to detect the center of the scene,and the omnidirectional image was transformed around the center. An adaptively updated background was modeled by the mixture Gaussian model for the transformed images. The moving object can be effectively segmented by the foreground segmentation method. The real time performance of target detection can be improved by adjusting the sample frequency during image transforming and mixture Gaussian modeling. Experimental results demonstrate that the system can effectively detect the moving object under complex environments, and has more accuracy and robustness.