【摘要】 为了解决多摄像机下矿井危险区域目标行为检测失效的问题,提出基于目标区域的配准算法.采用Lucas-Kanade光流法和基于块的背景运动补偿分别实现静态和复杂背景下的目标区域分割;用DOG对目标区域做尺度空间极值检测获得特征点对,并用主成分分析和尺度不变特征变换(PCA-SIFT)描述子作特征区域描述,在定义的目标区域匹配准则下,通过目标区域匹配度量比较实现匹配并通过基本矩阵约束消除误配区域.结果表明:该算法能够快速有效的实现矿井危险区域(低照度、目标类型复杂)下多摄像机运动目标匹配;与全图做PCA-SIFT匹配算法比较,计算复杂度降低71%～72%.更多还原

【Abstract】 To resolve the target-behavior detection in danger zones of coal mine based on multiple cameras,a novel object matching algorithm has been proposed based on object-regions.The object-regions were determined by the Lucas-Kanade optical flow algorithm and background motion compensation algorithm based on blocks under still and complex background.The scale invariant features of object-regions were obtained by a Difference-of-Gaussian (DOG) algorithm.The principal components analysis and scale invariant feature transform (PCA-SIFT) descriptors were selected as description of object-regions.Base on the defined matching criterion,the object-regions matching was achieved by comparison of matching measurements.The false matches in object-regions were eliminated using a fundamental matrix.The results show that the matching algorithm can deal with object matching in coal mine(low illumination level,complex target type) based on multiple cameras.The computational complexity of this object-regions matching algorithm was decreased by 71%-72% compared with PCA-SIFT matching algorithm in full image.更多还原