Please use this identifier to cite or link to this item: https://hdl.handle.net/10356/157722
Title: Visual recognition using artificial intelligence (person detection and tracking using artificial intelligence)
Authors: Shee Thoo, Sean Jun Hao
Keywords: Engineering::Electrical and electronic engineering
Issue Date: 2022
Publisher: Nanyang Technological University
Source: Shee Thoo, S. J. H. (2022). Visual recognition using artificial intelligence (person detection and tracking using artificial intelligence). Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/157722
Abstract: This report motive is to produce a robust real-time CCTV monitoring system aid to assist security guards, using only input video footage an object detection machine learning model and a tracking algorithm. The objective of this report is to investigate and determine which object tracking machine learning model and tracking algorithm is best for the task. The report presents the precision, recall, and F1 score of different versions of the state-of-the art object detection model family, You Only Look Once (YOLO), and the precision, swapping of track identification rates, new track identification rates, and speed of the state of-the-art multiple object-tracking algorithm family, Simple Online and Realtime Tracking (SORT). Different object detection machine learning models were trained and validated on 60,796 images of people in different angles, background, field of view and different scenarios. Once the best object detector was determined it was combined with different tracking algorithms and benchmarked on a labelled testing dataset consisting of 697 images/frames. The results of the research shows, in terms of detection YOLOv5-XLarge had the best detection performance while YOLOv5-Nano had the fastest speed. However, for a real-time CCTV monitoring system a balance between the detection performance and speed is key hence YOLOv5-Small is the most suitable for the task. In terms of tracking, Deep SORT with osnet_x0_5 has the best average precision while SORT has the best speed. While SORT has the best speed, Deep SORT with osnet_x0_5 speed is comparable to SORT but with a much larger precision value and consistency of tracking. Hence Deep SORT with osnet_x0_5 would be chosen as the tracking algorithm. Hence to produce a CCTV monitoring system to aid security guards, object detection from YOLOv5-Small Machine Learning Model would be combined with tracking using Deep SORT with osnet_x0_5 tracking algorithm.
URI: https://hdl.handle.net/10356/157722
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
Visual Recognition Using Artificial Intelligence.pdf
  Restricted Access
2.39 MBAdobe PDFView/Open

Page view(s)

48
Updated on Dec 2, 2022

Download(s)

5
Updated on Dec 2, 2022

Google ScholarTM

Check

Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.