SBIR Phase II: Large-Scale Behavioral Analysis Utilizing Convolutional Neural Networks and Its Application to In-store Retail Marketing

这个小企业创新研究（SBIR）第二阶段项目的更广泛的影响/商业潜力将通过暴露以前不可用的数据集来增强研究和教育的基础设施：细粒度的人类与物理环境的交互。人类总是在构建和塑造世界，但几乎没有硬数据来进一步研究这一影响。除了零售，这项技术还可以影响教师如何布置教室，灾难工作者如何提供救援，或者工厂如何保证工人的安全。如果没有拟议的系统，人们将以不可能的方式理解和研究每天影响人类的微妙物理细节。这项技术将通过提高经济效率造福社会，因为零售商可以更容易地满足客户的需求。这转化为巨大的潜在商业价值，因为零售市场的规模，因为零售商面临着提供差异化客户体验的压力，而客户无法在网上或大型商店获得这些体验。通过更深层次地了解客户，从而实现这种体验，而这反过来又是通过这里提出的技术实现的。这个小型企业创新研究（SBIR）第二阶段项目使用传感器放置模型，统计方法和人脸识别来充分实现第一阶段开发的视频分析系统的商业潜力，以了解人类行为。由于深度学习，基于视频的行为识别研究再次令人兴奋，但目前的工作只解决了问题的一部分。关键的缺失组件是强大的人脸识别和注册，安装尽可能少的摄像头同时最大限度地增加商店的可视区域的过程，以及与零售交易等其他数据相结合时有意义的视频分析结果。从视频中提取生物特征数据（如面部特征），自动计算摄像头的最佳位置，以及将行为指标与业务运营相关联的研究目标对于改善购物者的零售体验并使第一阶段的开发与该技术的潜在最终用户更具商业相关性至关重要。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase II project will be to enhance infrastructure for research and education by exposing a previously unavailable dataset: fine-grained human interaction with a physical environment. Humans are always building on and shaping the world but there is little hard data to further examine the effects this has. Beyond retail, this technology could affect how teachers layout classrooms, how disaster workers provide relief, or how factories keep their workers safe. The subtle physical details that affect humans everyday will be understood and investigated in ways not possible without the proposed system. This technology will benefit society by increasing economic efficiency because retailers can meet customer needs more easily. This translates into large potential commercial value because of the size of the retail market and because retailers are under pressure to deliver differentiated customer experiences that customers cannot get online or at big-box stores. Such experiences are enabled by understanding the customer at a much deeper level which in turn is enabled by the technology proposed here.This Small Business Innovation Research (SBIR) Phase II project uses sensor placement models, statistical methods, and face recognition to fully realize the commercial potential resulting from the Phase I development of a video analytics system for understanding human behavior. Research in video-based behavior recognition has seen renewed excitement because of deep learning but current work only addresses pieces of the problem. Critical missing components are robust face identification and registration, a process to install as few cameras as possible while maximizing the viewable area of a store, and video analysis results that are meaningful when combined with other data such as retail transactions. The research objectives of extracting biometric data (such as facial features) from video, automatically computing the optimal positions of cameras, and correlating behavior metrics with business operations are essential to improve the retail experience for shoppers and make the developments undertaken in Phase I more commercially relevant to potential end users of the technology.