COVID-19 Announcement Check our new UV-Drone project for home and workplace disinfection in distancing

Throughout history, the bird has been man's best friend. Indeed, the metaphor of an assistive flying creature, equipped with eyes, ears, and a nose to expand human senses goes back centuries and remains a prevalent motif of 21st century science-fiction pop culture like Peter Pan's TinkerBell, fueling a myriad of new drone-based consumer technology. As the Cheyenne Native American legend goes, the hawk's guidance allowed the once hunted and inferior human race to gain the respect of all animals and to climb to the top of the food chain. Our soaring bird is just the same. Well, almost. DetecDrone utilizes state of the art algorithms and solutions in computer vision, active learning, and data-driven optimization and control methods on various drone platforms in order to augment, expand, and enhance our human senses and cognitive abilities in various scenarios. From elderly care to precision livestock management to search and rescue, DetecDrone provides an intelligent unmanned aerial platform for sensing and information acquisition, and analytics. Once amplifying human dominance, our fellow flying friends now amplify human senses like never before. Explore our webpage for demos, projects, and publications showcasing DetecDrone.

Project Overview

An overarching capsulization of DetecDrone's unique machine learning mechanism.

In contrast to classic machine learning (ML) paradigm where inference and learning is conducted in the passive setting, drones collect data in an online fashion that is closely correlated to their flight path, movement, and speed. In other words, instead of learning from a large dataset consisting of measurements passively collected prior to the model learning and training, learning on drone platforms require acquiring information in an active fashion and via optimized and controlled sensing. A cohesive theoretical framework guiding the design and characterization of augmented controlled sensing and information acquisition is still lacking. We attribute this shortcoming in the current state of knowledge in the related fields of information theory, statistics, and machine learning to a lack of understanding regarding the 1) the temporal value of information as well as 2) the operational (task-specific) value of information. In contrast, our DetecDrone technology builds on our prior theoretical work in the general areas of active machine learning, controlled sensing, and information acquisition to design and operate an intelligent drone platform.