动态背景下基于帧间差分与模板匹配相结合的运动目标检测

基于图形处理器单元（GPU）提出了一种帧间差分与模板匹配相结合的运动目标检测算法。在CUDA—SIFT（于统一计算设备架构的尺度不变特征变换）算法提取图像匹配特征点的基础上，优化随机采样一致性算法（RANSAC）剔除图像中由于目标运动部分产生的误匹配点。运用背景补偿的方法将静态背景下的帧间差分目标检测算法应用于动态情况，实现了动态背景下的运动目标检测，通过提取目标特征与后续多帧图像进行特征匹配的方法最终实现自动目标检测。实验表明该方法对运动目标较小、有噪声、有部分遮挡的图像序列具有良好的目标检测效果。

Moving target detection combined two frame differences with template matching methods under dynamic background

A moving target detection algorithm is presented which combined frames difference with template matching methods based on Graphic Processing Unit(GPU).The method firstly utilizes CUDA-sift(a fast SIFT algorithm based on Compute Unified Device) algorithm to extract feature point of the matching image.Then,Random Sample Consensus(RANSAC) is optimized to remove the false matching points due to the part of the moving target.Put frames difference method under static background into dynamic setting using backdrop compensation method.Achieved the moving target detection method under dynamic background,extracting the feature of target to process character mating with the follow-up graphic finally realize automatic target detection.Experiments show that the method has favorable result of target detection for small moving target,little noise,some occlusion of image sequence.