红外技术, 2015, 37 (5): 404, 网络出版: 2015-06-11
基于超像素时空显著性的运动目标检测算法
Moving Target Detection Algorithm Based on Superpixel Spatiotemporal Saliency
摘要
针对复杂背景下运动目标检测存在的背景干扰、目标分割不完整等问题,利用目标静态灰度特征和运动特征,结合目标运动连续特性,提出了一种基于超像素时空显著图的运动目标检测算法.首先对图像基于简单线性迭代聚类算法(SLIC)进行超像素分割,以初始超像素为节点、以运动特征差异性为边建立图结构对超像素区域进行合并,得到最终超像素图像,可以有效解决传统超像素分割方法过分割而导致目标被分为多个部分的问题;然后分别利用目标静态特征对比度和运动特征对比度,得到静态显著性图和运动显著性图,并融合得到最终的时空显著性图;最后利用恒虚警处理技术,结合运动连续特性实现目标的检测,可以有效减少虚警目标.实验结果表明,该算法针对复杂背景具有良好的鲁棒性,并且可以比较完整的保留目标的信息.
Abstract
A new moving target detection algorithm based on superpixel spatiotemporal saliency is proposed to solve the problem of high false alarm and incomplete structure within moving target detection under complex background which takes into account of static characteristic and motion information of moving targets.Firstly,this algorithm extracts superpixels by image segmentation techniques,and builds a graph based on motion information and makes use of the graph theory for superpixels region emerging.Then,this algorithm calculates the static feature contrast and motion feature contrast of each superpixels to generate spatiotemporal saliency map.So the final saliency map can be obtained by fusing both the static and motion feature saliency map.Finally the target detection is achieved using constant false alarm processing and motion continuity feature.The experiment result shows that the algorithm is robust to complex background and detects intact information of the target.
云红全, 徐力, 孙骁, 明德烈, 鞠雯. 基于超像素时空显著性的运动目标检测算法[J]. 红外技术, 2015, 37(5): 404. YUN Hong-quan, XU Li, SUN Xiao, MING De-lie, JU Wen. Moving Target Detection Algorithm Based on Superpixel Spatiotemporal Saliency[J]. Infrared Technology, 2015, 37(5): 404.