Automated Object Detection and Classification using Metaheuristics with Deep Learning on Surveillance Videos

Abstract

Object detection and classification from surveillance videos is a procedure of exploiting computer vision (CV) and machine learning (ML) techniques for categorizing and identifying objects from realtime video streams. This assists in understanding and analyzing the scene, tracking object movement, and identifying potential security threats. This technology has been extremely utilized in security and surveillance schemes, autonomous vehicles, and retail analytics. This study develops an Automated Object Detection and Classification using Metaheuristics with Deep Learning on Surveillance Videos (AODC-MDLSV) technique. The presented AODC-MDLSV technique proficiently detects and classifies the objects into multiple classes. To do so, the AODC-MDLSV technique initially performs object detection using YOLO-v5 model. In the next stage, the AODC-MDLSV technique employs random vector functional link network (RVFL) method for object classification purposes. Finally, artificial bee colony (ABC) optimization methodology is used as a parameter optimization approach to improve the detection efficiency. The simulation values of the AODC-MDLSV approach are tested on benchmark video and the results showcased the better performance of the AODC-MDLSV method with maximum accuracy of 99.93%.