An Investigation of Context-Aware Object Detection based on Scene Recognition

Ayman Naushad, Radu-Casian Mihailescu

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Visual scene understanding for humans entails identifying regions of interest and then reasoning about information gained through their contextual analysis. Attempts to reproduce this process using Computer Vision techniques play an important role in improving the performance of Object Detection and general scene understanding algorithms. In this paper, we augment the training algorithm of an Object Detection model by integrating contextual information in the form of ‘scene labels’, through various methods to identify the superior approach. Several deep learning models were implemented and evaluated; results show that the Contextually Aware Object Detection Model performed the best, producing the highest classification accuracy of 75.14% and mean bounding box IoU of 0.69 on the constructed dataset. This was achieved by virtue of an auxiliary scene classification model used to make image scene predictions during the training phase.
Original languageEnglish
Title of host publication2023 International Conference on Modeling, Simulation & Intelligent Computing (MoSICom)
PublisherIEEE
Pages41-46
Number of pages6
ISBN (Electronic)9798350393415
DOIs
Publication statusPublished - 19 Mar 2024
Event2023 International Conference on Modeling, Simulation & Intelligent Computing - Dubai, United Arab Emirates
Duration: 7 Dec 20239 Dec 2023

Conference

Conference2023 International Conference on Modeling, Simulation & Intelligent Computing
Abbreviated titleMoSICom 2023
Country/TerritoryUnited Arab Emirates
CityDubai
Period7/12/239/12/23

Keywords

  • computer vision
  • context
  • object detection
  • scene recognition

ASJC Scopus subject areas

  • Artificial Intelligence
  • Energy Engineering and Power Technology
  • Electrical and Electronic Engineering
  • Hardware and Architecture
  • Computer Science Applications
  • Modelling and Simulation

Fingerprint

Dive into the research topics of 'An Investigation of Context-Aware Object Detection based on Scene Recognition'. Together they form a unique fingerprint.

Cite this