一种视频流特定人物检测方法

Specific Person Detection Method in Video Stream

为实现视频流特定人物检测,提出了一种两阶段检测方法,即基于稳健哈希签名的流级检测和基于子图局部特征的文件级检测。第一阶段利用签名快速发现与查询图像相似度较高的关键帧,第二阶段采用人脸子图的局部特征搜索与查询图像相似度次高的关键帧,它主要包括人脸检测、提取人脸子图的局部特征和相似计算三部分。实验结果证明该方法既可以快速检测到特定人物,也可以提升人物检测性能。

To detect a specific person in video stream, a two-step detection method is proposed, ie stream level detection based on robust Hash signature and file level detection based on local features of sub-image. The first step aims to find quite similar key frames to query image with image signature, and the second step is to find less similar key frame matching local features of face sub-image, including face detection, local feature extraction of face sub-image and similari- ty measure. The experimental results show that the method can detect specific person and im- prove the performance of human detection.