权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Deep Learning powered CCTV Monitoring

深度学习驱动的闭路电视监控

基本信息

批准号：
35540
负责人：
金额：
$ 29.08万
依托单位：
CALIPSA LIMITED
依托单位国家：
英国
项目类别：
Collaborative R&D
财政年份：
2019
资助国家：
英国
起止时间：
2019 至无数据
项目状态：
已结题

来源：
https://gtr.ukri.org/projects?ref=35540
关键词：
Deep Learning powered CCTV Monitoring

项目摘要

Companies such as G4S and Securitas started supplying human guarding services over 100 years ago to protect high-value sites from burglars. Gradually, human guards have been succeeded by CCTV cameras watched by human operators working in remote monitoring stations. With the widespread adoption of CCTV, particularly in the UK, these operators have been tasked with monitoring ever more video feeds, making it almost impossible for them to spot every potential crime. Calipsa engineers cutting-edge technology to solve this problem. The company's systems harness deep neural networks to analyse motion-triggered alarms, separating out false alarms from real suspicious events so that operators can review footage in real time and respond accordingly. These systems are highly successful at reducing the number of alarms that an operator has to view, helping them focus on the events that truly matter. This project will push the technology even further, automatically annotating objects in feeds so that operators don't have to spend time deducing what caused the alarm in often obscured footage or complex scenes, and bringing together feeds from multiple cameras capturing an event so that operators can follow the action easily. These innovations will deliver more in terms of function, accuracy and scale while working within the bounds of the current CCTV infrastructure.

G4S和Securitas等公司早在100多年前就开始提供人类守卫服务，以保护高价值遗址免受窃贼的侵扰。逐渐地，人类守卫已经被远程监控站工作的人工操作员监控的闭路电视摄像机所取代。随着闭路电视的广泛采用，尤其是在英国，这些运营商的任务是监控越来越多的视频源，使得他们几乎不可能发现每一起潜在的犯罪行为。哈里普萨设计了解决这一问题的尖端技术。该公司的系统利用深度神经网络来分析运动触发的警报，将虚假警报与真正的可疑事件区分开来，以便操作员能够实时查看视频并做出相应反应。这些系统非常成功地减少了操作员必须查看的警报数量，帮助他们专注于真正重要的事件。该项目将进一步推动这项技术的发展，自动对视频源中的对象进行注释，这样操作员就不必花费时间在经常模糊的镜头或复杂场景中推断是什么导致了警报，并将捕捉事件的多个摄像头的视频源结合在一起，以便操作员可以轻松跟踪操作。这些创新将在功能、精度和规模方面提供更多，同时在当前CCTV基础设施的范围内工作。