Organizations like the Department of Defense and the US Intelligence Community are continuously looking for ways to improve methods of gathering intelligence with the goal of improving public safety. Having human agents out in the field to gather data in-situ is expensive, not just in monetary terms, but also in time and human costs. One of the ways that these costs can be alleviated is to employ remote sensing through video surveillance. Coupled with object-tracking algorithms, this system would be able to provide real-time information in an area of interest. Being able to detect and track individuals autonomously will provide great benefits, with the aim of being able to detect threats and potentially avoiding a catastrophic event. Our particular research objective is to develop algorithms capable of real-time human detection in an area of interest from standalone video processing systems, as well as extracting tracks data. We are developing computer vision models for detecting humans that can be leveraged for other stationary and moving objects, such as vehicles, animals, and related objects. Our preliminary method of object detection has been to use a HAAR cascades and Local Binary Patterns as image features to train a machine-learning algorithm. However we are finding that these image features have limited accuracy in detecting our targets, namely human body shapes under challenging conditions like variation in lighting, color patterns and pose of the body, along with occlusion. Our next goal is to apply deep neural networks to improve the detection of human targets. Once we have the ability to reliably identify a target, we will begin working on a tracking algorithm.
A nother child page on the panel page template.