论文部分内容阅读
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。
In this paper, aiming at long-range target detection in visible and infrared thermal image sequences, a target detection method based on decision-level fusion is proposed. In this method, the target detection of each sensor image is first carried out through the method of accumulating and extracting the local grayscale information between the frames, and then the target detection result of each sensor is fused by the AND logic to remove part of the redundant information. Then, The multiple image features of each candidate region in the fusion detection result are extracted from each sensor image as evidence to further eliminate the redundant information. Finally, DS evidence theory is used to perform the target fusion recognition processing based on multiple features for each candidate region and the recognition result is taken as The ultimate goal of the entire system to detect output. Experimental results show the effectiveness of the proposed method.