Implementation of Pedestrian Tracking in Low-Resolution Video using Multi-Camera

Abstract

The utility of intelligent CCTV has been verified in many social domains. One of the main purposes of intelligent CCTV is pedestrian tracking. In this paper, we concentrate on tracking pedestrians using multiple cameras. Especially, we present a method that not only recognizes pedestrians from high-resolution camera views but also continuously tracks the recognized pedestrians from low-resolution camera views. Deep learning-based object detection model(YOLO-v4) and multi-object tracking algorithm(DeepSORT) are used in the suggested method to track pedestrians. Pedestrians are matched from the views of high-resolution and low-resolution cameras using a perspective transformation.