Design of real-time safety behavior monitoring and violation identification system for construction sites combining YOLOv8-Seg and DeepSORT

The existing detection framework systems are prone to misallocation in high-density environments and are difficult to adapt to deployment settings with limited network connections and diverse security standards. Therefore, this paper proposes an end-to-end real-time security monitoring system designed specifically for construction sites in network constrained environments. The system integrates four core modules: YOLOv8-Seg (You Only Look Once version 8-segmentation variant) for pixel-level instance segmentation; DeepSORT (Deep Simple Online and Real-time Tracking) for cross-frame identity tracking; Mask IoU (Intersection over Union) for object attribution assignment; and a sliding window -based cumulative violation detection mechanism. In addition, the system adopts INT8 (8-bit integer quantization) quantization for edge deployment to adapt to the network constraint environment of overseas sites. The experimental results show that compared with the baseline, the individual instance segmentation performance (mAP@0 .5, mean average precision) and cross frame identity association score (IDF1, identification F1 score) of the system show significant improvement. This system provides a real-time security management solution that meets international standards and offers a comprehensive chain of evidence for construction sites under network restrictions.