复杂背景下H.264压缩域运动目标检测算法

Moving objects detection method based on H.264 compressed video with complex background

针对H.264/AVC压缩码流中的运动目标检测问题,提出了一种基于马尔可夫随机场的最大后验概率(MAP-MRF,maximum a posterior-Markov random field)框架下适合复杂背景的H.264压缩域运动目标检测算法。算法首先生成滤波后的4×4像素块均匀运动矢量(MV,motion vector)场,对MV的相位建立高斯混合模型(GMM,Gaussian mixture model),结合MV幅度、帧间宏块分割模式、MV相位背景模型和运动目标时空约束建立马尔可夫随机场(MRF,Markov random field)模型。通过求解该模型得到每个4×4像素块前景、背景标记检测出运动目标。实验结果表明,算法能从复杂场景的H.264码流中提取出运动目标,与传统算法相比,Precision和Recall指标平均分别提高了20%和3.5%。

For the purpose of abstracting moving objects from H.264/AVC bit stream directly,a moving objects detection algorithm on H.264 compressed video with complex background which based on MAP-MRF framework was proposed.Firstly,it retrieved the moving vectors（MV） and the inter-prediction modes of identical 4×4 pixels block in P frames and establishes Gaussian mixture model（GMM） of the phase of MVs as a background,and then created Markov random field（MRF） model based on MV,inter-prediction mode,the GMM of background and spatial and temporal consistency.The moving objects would be retrieved by solving the MRF model.The experimental results show that it can perform robustly in complex environment and the precision and recall has improved 20% and 3.5% by contrast with the traditional algorithm.