摘要
在非受控环境中，由于背景的动态变化或光照、阴影的影响，执行高效、实时的运动目标检测具有很大的挑战性，联合长波红外（LWIR8～14gm）和可见光相机构成一个多模视觉系统可以显著提高运动目标检测的鲁棒性和完整性。提出了一种先检测后融合的运动目标检测算法，首先对可见光视频采用混合高斯建模方法检测运动目标，对热红外视频设计了基于背景差分和时间差分相结合的加权算法提取运动区域，然后对可见光与热红外视频中运动目标进行特征级融合。实验结果表明：该方法利用热红外与可见光图像的直观互补特征，在满足实时性要求的同时，可实现运动目标的精确、完整、鲁棒性检测。
In uncontrolled environments, because of the effect of dynamic background, lighting changes and shadows, it is challenging to perform an efficient and real-time moving target detection algorithm. Constructing a multi-mode visual surveillance system with long wave infrared （LWIR 8-14 m） and visible cameras can significantly improve the robustness and completeness of moving objects extraction. This paper presents a detection-fusion moving target detection algorithm. It starts from a Gaussian mixture background modeling algorithm for moving objects extraction in visible video and a weighted method based on background subtraction and the time-stepping for moving target detection in thermal video. The moving targets, obtained from visible and thermal video, are then fused at the feature level. The experimental results demonstrate that this method which uses the intuitive and complementary information from thermal and visual imagery can meet the real-time requirements, and can also get more complete, accurate and robust detection.
出处
《红外技术》
CSCD
北大核心
2013年第12期773-779,共7页
Infrared Technology
基金
西南科技大学研究生创新基金资助
中国电科集团公司CCD研发中心基础技术研究项目
西南科技大学网络融合工程实验室开放基金
