Quiet agent detection through simulation and classification

A central problem for organizations with a with a tactical surveillance mission is that of the “quiet agent.” The quiet agent problem concerns the real-time detection of a quiet entity moving in the midst of other entities, such as a silent person passing through an otherwise noisy crowd. The detection of the quiet agent is made possible by the effect it has on the surrounding agents and environment. In this paper we describe a proof-of-concept machine learning framework that is able to detect quiet agents in closed spaces where audio monitoring is available. Our approach begins by simulating an environment to produce relevant and usable data. The data is then converted into matrix form to be run through a neural network to detect the presence and movement of the quiet agent. The neural network was able to predict the location of the quiet agent with reasonable accuracy. Finally, the framework includes a step where data is normalized and heatmaps are generated, allowing the human eye to follow the quiet agent path.