WSLCC: A Weakly Supervised CNN-Transformer Model for Crowd Counting and Its Application in Sports Venue Management

Abstract

Aiming at the problem of low accuracy and poor adaptability of current crowd counting methods in sports venue management, an innovative crowd counting model based on weakly supervised learning (WSLCC) is proposed and a corresponding management platform is designed. In terms of model construction, this work combines weakly supervised learning ideas to deeply integrate traditional Convolutional Neural Networks (CNN) with Transformers. Firstly, advanced Convolutional Feature Module (CFM) is utilized to accurately capture and extract high-level semantic information of the crowd in video frames. Subsequently, this information is fed into an efficient Transformer Feature Module (TFM), which utilizes its powerful modeling capabilities to comprehensively construct global contextual information and long-range dependencies. Weakly supervised learning is reflected in using a small amount of labeled data to guide model learning and reduce dependence on a large amount of accurately labeled data. To validate the performance of the model, experiments are conducted on multiple datasets. The model ablation experiment shows that on the UCF_CC_50 dataset, the mean absolute error (MAE) of the WSLCC model is 62.8, and the root mean square error (RMSE) is 95.4, which is 2.5% and 0.2% lower than that of the LSC-CNN model, respectively. With the gradual addition of CFM and TFM modules, the model performance significantly improves, and the combined MAE and RMSE index values are the lowest. In practical applications, the sports venue management platform based on the WSLCC model achieves significant results, with an accuracy rate of 95.1% in crowd statistics, a venue utilization rate of 85.4%, a satisfaction score of 4.5 for resource allocation, and a management response time shortened to 5.3 minutes. This study effectively improves the adaptability and accuracy of crowd counting methods in complex environments, promoting the improvement of sports venue management efficiency.