Edge-Oriented Lightweight Facial Recognition for Real-Time Security Threat Detection in Sports Venues

To address the challenges of complex background handling, detail enhancement, and lightweight requirements when using facial recognition as a security screening tool in sporting events, a lightweight network is proposed for rapid face recognition. In this network, the integration of the GhostNet block with the Squeeze-and-Excitation block is used to reduce feature redundancy and computational costs while enhancing foreground target discrimination and suppressing background interference. The network is further configured to incorporate Cross Stage Partial Networks, a development which has the effect of reducing computational costs by means of the division of gradient flows, whilst retaining multi-scale feature representation capabilities. The model that was trained with this network was then tested for face recognition using public datasets. Experimental results demonstrate that the model attains an [email protected] of 91.8% in face recognition, with a frame rate of 9.5 FPS and a latency of 38.1 ms on edge devices, surpassing comparable models. The proposed model demonstrates outstanding effectiveness in real-world event scenarios, providing valuable technical insights for improving face recognition in sports event management.