Comparison of Object Detection Algorithms for the Task of Person Detection on Jetson TX2 NX Platform

Abstract

This article is an investigation for the best suited algorithm for object detection on Jetson TX2 NX platform. Previous research aimed at determination of the quickest and the most accurate algorithm for obj ect detection and tracking of vehicles. Current research uses previous article results and tries to apply found models on Jetson TX2 NX platform. The general goal is detection of a person on crosswalk. The model should have higher speed, than better accuracy in this case, because detection speed is a crucial point. This research includes applying of the YOLOv8m with 18 classes, which may appear on the road, YOLOv8s with only person class, DetectNet_v2 with ResNetl0 as a backbone and smaller version of this model. Mentioned models will be compared using trtexec tool on Jetson TX2 NX.