Cognitive microwave radar for the stand-off detection of on-body concealed weapons - phase IIb (partnership with a canadian company)

犯罪和恐怖主义行为在现代社会中备受关注。探测身体隐藏的武器对拯救生命至关重要。即使被扫描的人合作，目前的检测技术也存在很高的误判率，无论是阳性还是阴性。高成本和低吞吐量阻碍了在学校、体育场等繁忙的公共建筑中部署。隐蔽扫描在人们不知道自己正在被扫描的情况下，甚至更具挑战性，仍然是一个悬而未决的问题。然而，有时需要这种“隐蔽”系统来充分保护平民。该项目的重点是开发一种利用微波雷达对隐蔽武器进行实时扫描和检测的新产品。研究将集中在软件定义的无线电平台上，该平台在开发、部署甚至操作阶段提供极大的灵活性。它允许在不更改硬件的情况下改进系统，并通过安全的网络连接进行更新。这种体系结构可以作为芯片上的系统来使用，这将减少最终用户的体积和成本。该系统的硬件占地面积将大致相当于一本笔记本的大小。由于辐射是非电离和低功率的，因此检测系统是安全的。该产品的另一个优势是，不会生成身体图像而会检测到被测量的物体，从而避免了隐私问题。该产品的一个显著特征是它具有不断了解部署地点、新威胁(如武器)和新的非威胁(如手机)的认知能力。识别是通过机器学习分类器实现的，机器学习分类器通过对威胁、非威胁和背景环境的大集合雷达响应进行统计分析来进行“训练”。分类在很大程度上取决于这些测量，特别是部署地点的测量。因此，产品在安装后将继续获得新的测量结果，以提高检测成功率。还将通过安全网络更新提供新的威胁/非威胁数据以及新的分类器。

Crime and acts of terrorism are of great concern in modern society. The detection of on-body hidden weaponsis of paramount importance for saving lives. Current detection technologies suffer from high false rates, bothpositive and negative, even if the scanned people cooperate. High cost and low throughput prevent deploymentin busy public buildings such as schools, stadiums, etc. Covert scanning, where people are unaware that theyare being scanned, is even more challenging and remains an unsolved problem. Yet such "covert" systems aresometimes needed to adequately protect civilians.This project focuses on the development of a new product for real-time scanning and detection of concealedweapons using microwave radar. Research will focus on a software-defined radio platform that offers greatflexibility during development, deployment and even operational phases. It allows for system improvementswithout hardware changes and for updates over secure network connections. This architecture is available as asystem on a chip, which will reduce the size and the cost to the end-user. The system's hardware footprint willbe roughly the size of a notebook.The detection system is safe as the radiation is non-ionizing and low-power. Another advantage is that theconcealed objects will be detected without generating an image of the body, thus avoiding privacy concerns.A distinct feature of this product is its cognitive ability to continuously learn about the deployment site, aboutnew threats (e.g. weapons), and about new non-threats (e.g. cell phones). Recognition is enabled by machinelearning classifiers, which are "trained" by applying statistical analysis to large sets of radar responses tothreats, non-threats, and background environments. Classification depends critically on these measurements,particularly those of the deployment site. Therefore, the product will continue to acquire new measurementsafter installation in order to improve the detection success. New threat/non-threat data, as well as novelclassifiers, will also be provided via secure network updates.