ceceliar630991/item-tracking-card2960
1
0
Code Issues Actions Packages Projects Wiki

Add Fast and Resource-Efficient Object Tracking on Edge Devices: A Measurement Study

Cecelia Nickerson 2026-01-13 05:12:01 -05:00
commit 4ada39d108
1 changed files with 5 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
Fast and Resource-Efficient Object Tracking on Edge Devices%3A A Measurement Study.-.md Normal file

@ -0,0 +1,5 @@
<br>Object monitoring is a crucial performance of edge video analytic programs and providers. Multi-object monitoring (MOT) detects the shifting objects and tracks their locations body by body as actual scenes are being captured into a video. However, it's well known that actual time object [item tracking card](https://opensourcebridge.science/wiki/Revolutionizing_Personal_Belongings_Protection_With_Tagsley_Tracker) on the edge poses important technical challenges, especially with edge units of heterogeneous computing sources. This paper examines the performance issues and edge-specific optimization alternatives for object tracking. We are going to show that even the properly educated and optimized MOT mannequin may still suffer from random frame dropping issues when edge gadgets have insufficient computation resources. We present a number of edge particular efficiency optimization methods, collectively coined as EMO, to hurry up the real time object monitoring, starting from window-primarily based optimization to similarity primarily based optimization. Extensive experiments on widespread MOT benchmarks exhibit that our EMO approach is competitive with respect to the consultant strategies for on-device object monitoring methods in terms of run-time efficiency and tracking accuracy.<br>
<br>Object Tracking, Multi-object Tracking, Adaptive Frame Skipping, Edge Video Analytics. Video cameras are widely deployed on cellphones, autos, and highways, and are soon to be out there nearly in all places sooner or later world, together with buildings, streets and varied kinds of cyber-physical methods. We envision a future the place edge sensors, reminiscent of cameras, coupled with edge AI providers might be pervasive, [item tracking card](http://local315npmhu.com/wiki/index.php/User:LEZYanira5) serving as the cornerstone of good wearables, smart properties, and good cities. However, a lot of the video analytics today are sometimes performed on the Cloud, which incurs overwhelming demand for network bandwidth, thus, transport all the videos to the Cloud for video analytics is not scalable, not to mention the different types of privateness issues. Hence, actual time and useful resource-aware object tracking is a crucial performance of edge video analytics. Unlike cloud servers, edge units and edge servers have restricted computation and communication useful resource elasticity. This paper presents a scientific examine of the open research challenges in object tracking at the sting and the potential efficiency optimization opportunities for fast and resource efficient on-gadget object monitoring.<br>
<br>Multi-object monitoring is a subgroup of object tracking that tracks multiple objects belonging to a number of classes by identifying the trajectories because the objects transfer through consecutive video frames. Multi-object monitoring has been widely applied to autonomous driving, surveillance with security cameras, and activity recognition. IDs to detections and [Tagsley tracking card](https://haderslevwiki.dk/index.php/Brugerdiskussion:SelmaLorenzini) wallet [Tagsley tracker](https://trevorjd.com/index.php/Revolutionizing_Tracking_Technology:_A_Case_Study_Of_Tagsley) tracklets belonging to the same object. Online object monitoring aims to process incoming video frames in actual time as they're captured. When deployed on edge devices with useful resource constraints, the video body processing fee on the sting machine may not keep pace with the incoming video body rate. On this paper, we give attention to reducing the computational price of multi-object tracking by selectively skipping detections whereas still delivering comparable object tracking quality.