Article: Abnormal swimming behaviour detection and multi-object tracking based on YOLOv7 and DeepSORT Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.52 pp.20 - 40 Abstract: With the development of society and the improvement of people's living standards, swimming and fitness have gradually become important activities in daily life. To improve the safety management of swimming facilities, this study proposes a YOLOv7 and DeepSORT algorithm for abnormal swimming behaviour detection and multi-object tracking. This method first uses YOLOv7 for object detection, and then continuously tracks the detected targets through the DeepSORT algorithm. To optimise feature extraction for small targets, this study utilises spatial pyramid deformable convolution module and non-local attention module attention mechanism for improvement. In addition, to further improve tracking accuracy, DeepSORT introduces distance intersection and union ratio. The results showed that the improved object detection accuracy, recall and F1-value reached 94.56%, 93.89%, and 95.08%, respectively. The accuracy of multi-object tracking on the training and testing sets reached 88.56 and 90.54, with an improved accuracy value of 89.42. In addition, the detection rate of the research method exceeded 86% in crowded scenes and 91% in sparse scenes, with a minimum false alarm rate of only 1.2 times per hour. The constructed method can identify and track abnormal swimming behaviour, providing technical support for pool safety management. Inderscience Publishers - linking academia, business and industry through research

Title: Abnormal swimming behaviour detection and multi-object tracking based on YOLOv7 and DeepSORT

Authors: Ting Li

Addresses: Sports Department, University of Shanghai for Science and Technology, Shanghai, 200093, China

Abstract: With the development of society and the improvement of people's living standards, swimming and fitness have gradually become important activities in daily life. To improve the safety management of swimming facilities, this study proposes a YOLOv7 and DeepSORT algorithm for abnormal swimming behaviour detection and multi-object tracking. This method first uses YOLOv7 for object detection, and then continuously tracks the detected targets through the DeepSORT algorithm. To optimise feature extraction for small targets, this study utilises spatial pyramid deformable convolution module and non-local attention module attention mechanism for improvement. In addition, to further improve tracking accuracy, DeepSORT introduces distance intersection and union ratio. The results showed that the improved object detection accuracy, recall and F1-value reached 94.56%, 93.89%, and 95.08%, respectively. The accuracy of multi-object tracking on the training and testing sets reached 88.56 and 90.54, with an improved accuracy value of 89.42. In addition, the detection rate of the research method exceeded 86% in crowded scenes and 91% in sparse scenes, with a minimum false alarm rate of only 1.2 times per hour. The constructed method can identify and track abnormal swimming behaviour, providing technical support for pool safety management.

Keywords: YOLOv7; DeepSORT; behaviour detection; multi-object tracking; swim.

DOI: 10.1504/IJICT.2025.151165

International Journal of Information and Communication Technology, 2025 Vol.26 No.52, pp.20 - 40

Received: 31 Jul 2025
Accepted: 30 Oct 2025
Published online: 15 Jan 2026 *

Title: Abnormal swimming behaviour detection and multi-object tracking based on YOLOv7 and DeepSORT

Keep up-to-date