Unmanned Aerial Vehicles Identification and Tracking Based on Video Data Analysis

Abstract

The proposed method implements automatic capture, tracking and identification of unmanned aerial vehicles (UAVs) in video sequence frames under the following conditions. The scene has a moving background, which in general can be highly detailed (complex background), or smooth; the size of objects of interest from 8×8 to 100×100 pixels; speed of movement from zero to 10 elements per frame; several objects move within the frame, including the object of interest; objects can be both artificial and natural. The proposed method at the stage of segmentation and tracking allows to implement segmentation for objects of interest with a size up to 5 blocks with a random error less than 15% and for objects of larger sizes 3-5%. Systematic error less than 13%. At the identification step, the balanced accuracy metric is 0.83. The novelty of the presented work is that for the first time the design approach and research results of a comprehensive UAV detection, tracking and identification procedure for various surveillance conditions and with a significant change in the size of the object in the frame during tracking are described.