基于超像素分割和混合权值AdaBoost运动检测算法  

AdaBoost Moving-target Detection Algorithm Based on Superpixel Segmentation and Mixed Weight

作　　者：李忠海 杨超 梁书浩 

机构地区：[1]沈阳航空航天大学自动化学院,沈阳110136

出　　处：《电光与控制》2018年第2期33-37,78,共6页Electronics Optics & Control

基　　金：航空科学基金(XY201434-2)

摘　　要：针对长时间目标跟踪检测不准确问题,提出一种结合运动场景的超像素分割与混合权值的Ada Boost多目标检测(ABSP)算法。首先在动态模型中,计算Ada Boost算法的混合权值,检测运动目标,确定搜索区域,提高多目标跟踪检测能力;在训练阶段,采用SLIC分割与Mean-Shift聚类形成超像素图块,构建目标外观模型;在跟踪阶段,结合超像素特征池生成模板直方图与置信图,构建观测模型与运动模型,采用粒子滤波与贝叶斯模型,计算最大后验估计,实现遮挡运动目标检测。结果表明:能够有效处理数目变化多目标检测与遮挡问题,提高了检测的实时性。In order to solve the inaccurate detecting problem in the long-time target tracking process, an AdaBoost multi-target detection algorithm is proposed based on superpixel segmentation and mixed weight. In the dynamic model, the mixed weight of AdaBoost algorithm is calculated out, the moving targets are detected, and the search area is determined, so as to improve the multi-target tracking and detecting capabilities. At the training stage, the superpixel segment is formed by using the SLIC segmentation and the Mean-Shift clustering, and the appearance model of the targets is built. At the tracking stage, the histogram and the confidence map of the template are created by using the superpixel feature pool, and the observation model and the motion model are built. The maximum posterior estimation is computed by using the particle filter and Bayes model, so as to realize the detection of sheltered moving targets. Experimental results show that： The proposed algorithm can effectively deal with the problems of varying-number, multi-target detecting and sheltered target detecting, and improve the real-time performance of the detection.

关 键 词：目标跟踪 多目标检测 超像素 AdaBoost检测 外观模型 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]