摘要： 为在视频流上实现特定人物实时检测，提出基于稳健哈希签名的人物检测方法. 首先设计稳健哈希函数并计算目标图像的签名，然后在实时视频流上计算关键帧图像的签名，再衡量两者相似程度并判断该视频是否含有设定的目标. 该方法能实时发现与设定人物相似的视频关键帧，从而定位到该关键帧所在的镜头，实现视频流特定人物检测功能. 实验表明该方法可达到较高的准确率，这对于及时获取特定信息有重要作用. 该方法也可以用于图像复制检测、重复视频检测等.

Detecting a specific person in video sequence is a challenging problem. We propose a real-time video detection method based on robust Hash signature. The approach consists of three steps. First, it devises the robust Hash function and computes the signature of the target image. Then, the signature of every representative key frame is acquired from the video sequence. Finally, the distance of two signatures is calculated to decide whether the target image is present. The proposed approach can capture key frames that may include a specific person in real-time, and locate the shot containing the detected key frames. Experimental results demonstrate that the approach can achieve high accuracy for retrieving repetitive specific information from video sequences, and can be used in applications such as detecting image copies and duplicate contents.