We are proud to announce, that we become a member of NVIDIA Inception Program! This virtual accelerator program led by NVIDIA nurtures dedicated and exceptional startup companies, who are revolutionizing industries with advances in AI and data science, and help them in all stages of product development, prototyping, and deployment. Using NVIDIA Pascal GPUs and Caffe’s deep learning framework, DataFromSky is able to do the automatic extraction of time-space-trajectories of moving objects in aerial video data captured by drones and/or balloons. The overall system is divided into three main parts: preprocessing, vehicle detection, and tracking. In the preprocessing step, the acquired image gets undistorted and geo-registered against a user-selected reference frame. The current solution for detection and tracking utilizes machine powered by NVidia GTX 1080 cards.
We designed a highly optimized multi-layer detector and trained it on a hand annotated training dataset with 10.2M positive and 10.2M negative samples per class. The detector is capable of learning from its errors and improving over time. Due to the outstanding performance of detection and tracking algorithms, DataFromSky can provide high-quality data and guarantee a hit rate of over 96%. See the raw output from DFS deep neural networks below: