Identify and Trace Criminal Suspects in the Crowd Aided by Fast Trajectories Retrieval

Abstract

Aided by the wide deployment of surveillance cameras in cities nowadays, capturing the video of criminal suspects is much easier than before. However, it is usually hard to identify the suspects only according to the content of surveillance video due to the low resolution rate, insufficient brightness or occlusion. To address this problem, we consider the information of when and where a suspect is captured by the surveillance cameras and achieve a spatio-temporal sequence ζ _i . Then we search the records of mobile network to locate the mobile phones which have compatible trajectories with ζ _i . In this way, as long as the suspect is carrying a mobile phone when he is captured by surveillance cameras, we can identify his phone and trace him by locating the phone. In order to perform fast retrieval of trajectories, we propose a threaded tree structure to index the trajectories, and adopt a heuristics based query optimization algorithm to prune unnecessary data access. Extensive experiments based on real mobile phone trajectory data show that a suspect’s phone can be uniquely identified with high probability while he is captured by more than four cameras distributed in different cells of the mobile network. Furthermore, the experiments also indicate that our proposed algorithms can efficiently perform the search within 1 second in the trajectory dataset containing 104 million records.