基于背景差分法和显著性图的海底目标检测方法

Underwater object detection based on background subtraction and a saliency map

如何完整地提取视频中低对比度的小目标一直是运动目标检测中的热点和难点,尤其是处理水下移动设备拍摄的海底视频,更是具有很大挑战性。针对海底视频的特点,提出一种结合背景差分法和显著性图理论的弱小目标检测方法。该方法对减去背景后仍不清晰的含有目标的视频帧,采用显著性图理论提高目标的对比度,以利于更好地分割目标与背景;同时,利用单帧视频中最大目标个数的先验知识,来判断视频中是否有目标出现,避免因杂质太多而造成目标的错检,以提高检测系统的稳定性。在此基础上进行了与常规背景差分法的比较实验,结果显示该方法能够更好地检测到不清晰的目标。

Detecting small low-contrast targets is a difficult hotspot,especially in underwater videos captured by moving remote-operated vehicles.According to the features of seabed videos,a small-targets detection method based on background subtraction and a saliency map was proposed.The saliency map was used to improve the contrast of targets that were still not clear enough after background subtraction and to make them easier to be segmented from the background.Whether or not the targets appeared in the video was determined with the prior knowledge about the largest number of subjects in a single-frame to avoid wrong detection and improve the stability of the detection system.Experimental results showed that unclear objects could be better detected with the proposed method than with conventional background subtraction.