Transformer-Based Multi-Object Detection Model for Smart Communities

Abstract

This paper proposes a Transformer-based multi-target detection model for smart community monitoring systems, integrating the YOLO algorithm. The model targets uncivilized behaviors such as littering from high-rise buildings, pet waste disposal, and illegal parking of electric vehicles, achieving instance segmentation, dynamic tracking, and real-time alerts. By introducing a Transformer module to enhance the attention mechanism, detection accuracy and robustness are improved; trajectory tracking and 3D model fusion are implemented, supporting multi-source data processing; and multimodal data fusion reduces the false alarm rate and supports humanistic care applications. Experimental results show that the proposed model significantly outperforms the benchmark YOLO in detection accuracy and real-time performance, validating the effectiveness of the innovation and providing practical technical support for improving community management efficiency.