Dynamic Target Tracking and Geospatial Transformation in Place-Based on DNN

Abstract

Video surveillance is critical for public safety. In open spaces such as urban places, how to conduct an intelligent analysis of surveillance video to obtain the dynamic target trajectory in the place is valuable for monitoring the behavior and motion situation of the dynamic target. According to the latest progress in the field of Deep Neural Network (DNN) and object detection and tracking, this paper aims to develop the method for dynamic Target Tracking and geographical transformation in the place. The detection model, YOLOv3 is used to extract dynamic target features, and dynamic target tracking is performed based on the DeepSORT method. The trajectories of the target are generated in the video frames and transformed into geographic space using the homography transformation method to visualize and analyze it based on GIS. As the experimental shows, the integrated DeepSORT and YOLOv3 models can quickly detect and track dynamic targets, and map target trajectories to geographic space, which could provide essential support for target trajectory analysis and motion situation awareness based on geospatial information in video surveillance.