摘要
本文针对可见光和红外热图像序列中远距离目标检测,提出了一种基于决策级融合的目标检测方法。该方法首先是通过帧间差累积和提取局部灰度信息的方法对各传感器图像进行目标检测处理;接着采用“与”逻辑对各传感器的目标检测结果进行融合,除去部分冗余信息;然后在各传感器图像中提取融合检测结果中各候选区域的多个图像特征作为进一步消除冗余信息的证据;最后采用D-S证据理论对各候选区域进行基于多特征的目标融合识别处理并将识别的结果作为整个系统最终的目标检测输出。实验结果证明了本文方法的有效性。
Aim at the problem of detecting for long distance targets with visual and thermal infrared image sequences, a method of targets detection based on muhisensor decision level fusion was developed. The algorithm firstly acquires targets detection from each sensor by frame difference accumulation and local intensity image; and then fuses the results using ＂and＂ logic to reduce the part of redundant information. Then extracts the multiple features of these target candidate areas in fusion detection result from the two sensor images as the evidence to eliminate redundant information. Finally, to distinguishe the false target from real target using D-S evidence theory based on multiple features, and send the recognition result to the target detection output of the whole system. The experimental results demonstrated that this approach is feasible and robust.
出处
《中国体视学与图像分析》
2007年第1期28-32,共5页
Chinese Journal of Stereology and Image Analysis
基金
国家重点实验室基金资助(No.51483040105QT5118)
关键词
可见光图像序列
红外图像序列
决策级融合
D-S证据理论
visual image sequences
thermal infrared image sequences
decision level fusion
D-S evidence theory
