Research on Multi-object Detection in Security Video Surveillance Based on Improved YOLOv8n
Abstract To address the issues of slow processing efficiency, low accuracy, and the difficulty in multi-object recognition of large video streams in the cloud, this paper optimizes the security video detection algorithm based on the YOLOv8n model. It introduces the SEAttention mechanism into the backbone network of the model to enhance the perception of key area features, adds a small object detection head (named P2) to provide higher-resolution feature maps for better detection of small objects, and adopts the WIoU V3 loss function to optimize the bounding box regression. To test the effect of multi-object recognition, the network data and on-site video slices are integrated, and after preprocessing, unified annotation, quality verification, and multi-dimensional enhancement, a multi-task dedicated dataset for fire smoke and safety helmets is constructed. The experimental results show that the improved YOLOv8n model has a precision rate of 86.5%, a recall rate of 77.5%, and an mAP50 of 84.3%. The core indicators have all improved compared to the original model, and the detection effect of small objects such as flames and safety helmets has been significantly improved. Moreover, the number of parameters and computational requirements are suitable for deployment on edge computing devices. This model effectively solves the real-time detection problem of fire smoke and safety helmets in complex environments, verifies its applicability in industrial security scenarios, and provides technical support for the intelligent security of industrial facilities.
Authors
- Tian Ze
- Jianing Liu
- Huiling Jiang
- Weiyi Min
- Jincheng Huang
Institutions
- University of Science and Technology Beijing (CN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-17
- DOI
- https://doi.org/10.21203/rs.3.rs-8559020/v1
- Primary Topic
- Advanced Neural Network Applications
- Type
- preprint