SGHM与MGN融合的跨镜识别算法及高空作业应用研究OA

To address the critical challenges in safety supervision for high-altitude operations,including visual blind spots,high false alarm rates,and response delays,an innovative cross-camera tracking(CCT)solution is proposed that integrates semantic guided human matting(SGHM)and multi-granularity network(MGN).The SGHM algorithm accurately segments human body regions,effectively eliminating interference from complex and dynamic backgrounds.Meanwhile,the MGN network innovatively employs parallel multi-granularity feature learning and a deviation-aware mechanism,endowing the model with inherent resistance to front-end segmenta-tion errors and fundamentally resolving feature discrimination issues caused by identical workwear,occlusion,and perspective changes.Experimental results demonstrate that the proposed solution achieves a Rank-1 accu-racy(Rank-1)of 95.3％and a mean average precision(mAP)of 96.5％on the Market-1501 dataset,and 94.0％Rank-1 with 88.8％mAP on the DukeMTMC-ReID dataset.On the self-built high-altitude operation dataset,the proposed algorithm attains a mean intersection over union(mloU)of 86.7％,an overall Rank-1 of 93.8％and mAP of 89.5％,significantly outperforming comparative methods.The ablation studies verify the effective-ness of both the SGHM preprocessing module and the dynamic feature distance loss function.Practical applica-tion cases confirm that the system enables effective identity tracking and behavior analysis for high-altitude workers,substantially enhancing the intelligence of safety supervision.The technical framework demonstrates broad application potential in high-risk fields such as construction,power and bridges.