Deep Learning powered CCTV Monitoring

G4S和Securitas等公司早在100多年前就开始提供人工护卫服务，以保护高价值地点免遭窃贼的盗窃。逐渐地，人工警卫被远程监控站的人工操作员观看的闭路电视摄像机所取代。随着闭路电视的广泛采用，特别是在英国，这些运营商的任务是监控越来越多的视频馈送，这使得他们几乎不可能发现每一个潜在的犯罪。Calipsa工程师采用尖端技术来解决这个问题。该公司的系统利用深度神经网络来分析运动触发的警报，将假警报与真实的可疑事件区分开来，这样操作人员就可以实时查看视频并做出相应的反应。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该技术的发展，自动标注馈送对象，这样操作人员就不必花时间在模糊的镜头或复杂的场景中推断是什么引起了警报，并将多个摄像机捕获的事件馈送到一起，这样操作人员就可以轻松地跟踪行动。这些创新将在功能、准确性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。

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.