Object Tracking Data Labeler

As part of my 2020 Summer Undergraduate Research Apprenticeship, I was tasked with implementing an "assisted video-labeler." To do so, I leveraged OpenCV's object tracking computer vision algorithms to automatically update bounding boxes given an initial box and a sequence of frames. This makes the process of labeling video data, specifically for the purpose of training an object detection Convolutional Neural Network (CNN), much faster and more intuitive, with several ease of use features incorporated into the software to make labeling as efficient as possible. This software was used by the SubT team, which required an efficient way to label different target objects for RGB-camera-based object detection and classification in a search and rescue scenario (e.g. gear, survivors, etc.). Our team achieved second place at the DARPA Subterranean challenge. Check out the GitHub repo for this project here.