基于步态识别的视频侦查技术与应用

Technology and Application of Video Detection Based on Gait Recognition

步态是远距离复杂视频监控场景下唯一可清晰成像的生物特征,具有识别范围广、难以伪装等优点。为探索视频图像智能化应用,发掘海量视频资源的潜在价值,开发了步态应用实战平台,利用步态识别技术破解视频监控条件下人员身份识别难和轨迹刻画难等难题,为侦查破案、人员管控等提供新手段。采用先进的深度学习步态识别方法,完成步态检测,序列跟踪,步态分割和识别比对等一系列关键环节和步骤。为验证步态识别技术的效果,在含10万段实战场景步态序列的数据库上测试,步态识别的top 1命中率可超过96%。在100万段实战场景步态序列的测试中,步态识别的top1命中率也可超过92%。该技术满足公安实战需求,为科技强警提供全新技术支撑,大幅提升视频侦查实战能力。

With the advantage of recognizing large-range targets and being difficult to disguise,gait is almost the only biometric feature for recognizing humans at a distance under complex surveillance scenes.To enhance the ability of video-based investigation and discover the value of large-scale video resources,we have developed a gait recognition platform to tackle the challenges of identifying and locating pedestrians appearing in real-world surveillance cameras.In this paper,the advanced deep learning based methods were used to complete the steps of gait detection,tracking,segmentation and recognition.Being tested on the gait retrieval task,our algorithm can achieve the top-1 accuracy of 96%on the 100,000 sequences database and 92%on a more challenging 100,000,000 sequences database.The results have shown that gait recognition can serve as a new tool to greatly facilitate the video-based investigation,which meets the application demands of the public security department.