Downloads: 7 | Views: 86 | Weekly Hits: ⮙1 | Monthly Hits: ⮙2

Research Paper | Computer Science and Information Technology | India | Volume 12 Issue 10, October 2023

Automated Object Detection and Classification using Krill Herd Algorithm with Deep Learning on Surveillance Videos

V. Saikrishnan | Dr. M. Karthikeyan ^[3]

Abstract: In the domain of video surveillances, the implementation of deep learning (DL) for object detection and classification is developed as a game-changer. This paper introduces a comprehensive solution integrating the power of DL methods to handle these fundamental tasks. Leveraging state-of-the-art neural networks (NN), our technique allows consistent object identification and categorization within surveillance videos, providing improved security, real-time context awareness, and enriched decision-making abilities. This study develops an Automated Object Detection and Classification using Krill Herd Algorithm with Deep Learning (AODC-KHADL) technique on Surveillance Videos. The introduced AODC-KHADL method efficiently detects and classifies the objects into numerous categories. This technique starts with the incorporation of YOLO-v5, a recent object detection method popular for its excellent accuracy and speed for identifying objects in videos and images. For enhancing YOLO-v5's detection potential, we utilize Random Vector Functional Link (RVFL) classification, a multipurpose and robust machine learning (ML) approach. In this context, we present the Krill Herd Algorithm (KHA), a nature-inspired optimization method inspired by the collective behavior of krill swarms. By using extensive examination and assessment, we exhibit the model's capability in real-time video surveillance applications. The simulation values of the AODC-KHADL technique are tested on benchmark video and it is emphasized the higher performance of the AODC-KHADL system with other models.

Keywords: Object classification; Video surveillance; Object detection; Krill Herd Algorithm; Deep learning

Edition: Volume 12 Issue 10, October 2023,

Pages: 86 - 92