Large-scale license plate analytics for convoy detection

该项目旨在开发用于分析车牌识别(LPR)数据流的计算方法。车牌识别摄像头安装在道路沿线或安装在执法车辆上，捕捉和记录对路过车辆牌照的观察(时间和位置)。总部设在蒙特利尔的Genetec，Inc.是向全球市场提供车牌识别技术的领先供应商，该项目建立在McGill和Genetec研究团队之间现有合作的基础上。该项目的目的是研究和开发统计算法，用于使用城市周围的摄像机集合记录的LPR事件流来检测一起行驶的车辆组(车队)。侦测和追踪车队的能力将有助于执法部门，例如，帮助抓捕偷车贼，或抓捕运输或走私非法货物的犯罪分子。这项任务的一个具有挑战性的方面是开发能够区分真正作为车队一起行驶的车辆组和由于交通模式而碰巧彼此靠近较长时间的车辆组(例如，在高峰时间或沿着一长段高速公路)。另一个挑战是开发能够扩展以处理非常大的LPR系统(数百台摄像机，每台摄像机每小时报告数千次事件)的算法。该项目将专注于结合稀疏信号处理、矩阵分解和优化的发展来开发可扩展的车队检测方法。我们还将考虑使用云中的计算机集群高效实施车队检测系统的任务，并将对矩阵因式分解的大规模和分布式优化的文献做出贡献。该项目开发的技术将转移给Genetec，并整合到他们的AutoVu产品中，该项目的成功可能会带来新的就业机会。通过Genetec的技术和服务，这些方法将提供给加拿大和世界各地的执法人员。

This project aims to develop computational methods for analyzing streams of license plate recognition (LPR) data. LPR cameras, positioned along roads or mounted on law-enforcement vehicles, capture and record observations (time and location) of the license plates of vehicles passing by. Based in Montreal, Genetec, Inc., is a leading provider of LPR technologies to the world-wide market, and this project builds on an existing collaboration between the research team at McGill and Genetec. The aim of this project is to study and develop statistical algorithms for detecting groups of vehicles traveling together (convoys) using streams of LPR events recorded by a collection of cameras around a city. The ability to detect and track convoys would be useful to law enforcement offices for, e.g., helping to catch car thieves, or for catching criminals transporting or smuggling illicit goods. One challenging aspect of this task is to develop algorithms that can distinguish between groups of vehicles that are truly traveling together as a convoy and vehicles which just happen to be near each other for an extended period due to traffic patterns (e.g., during rush hour, or along a long stretch of highway). Another challenge is to develop algorithms that are capable of scaling to handle very large LPR systems (hundreds of cameras, with each camera reporting many thousands of events per hour). This project will focus on developing scalable convoy detection methods combining developments in sparse signal processing, matrix factorization, and optimization. We will also consider the task of efficiently implementing a convoy detection system using a cluster of computers in the cloud, and contributions will be made to literature on large-scale and distributed optimization for matrix factorization. The technology developed in this project will be transferred to Genetec and integrated into their AutoVu product, and success of this project is likely to lead to the creation of new jobs. Through Genetec's technology and services, these methods will be available to law enforcement agents in Canada and around the world.