基于改进YOLOv12的智慧教室实时人数统计研究

打开文本图片集

中图分类号：TP391 文献标识码：A文章编号：1006-8228（2026）02-35-06

Abstract：Toadresstheproblemsoflargecomputation，highmemoryacessoverheadandpoorreal-timeperformanceinthe traditionalYOLOv12model，thispaperproposesanimprovedYOLOv1schemebasedonpartialconvolution（PConv）.Thecoreof thisschemeliesinleveragingthefeaturemapredundantinformationfiteringcapabilityofPConvcombinedwithanadaptie chanelalcationstrategytodynamicallyadjusttheproportionofefectiveconvolutionchannelsindiferentnetworklayers. Experimentalresultsshowthat，comparedwiththeoriginal YOLOvl2model，theproposedschemeachievesa 29.3% reduction in FLOPsand a 36.7% decrease in MAContheself-built dataset while maintaining detectionaccuracy.Thisenablesthe schemeto mettherequirementsofrealtimeclasroompeoplecoutingonlow-enddevices，therebyprovidingefectivetechnicalsupportfor intelligent education management and demonstrating certain application value.

Keywords：YOLOvl2；PConv;Real-time People Counting；Smart Classroom

0引言

为贯彻落实《教育强国建设规划纲要（2024—2035年）》，教育部把教育数字化作为关键突破口，通过开拓教育发展新赛道、打造发展新优势，全面助力教育强国建设，并出台《关于加快推进教育数字化的意见》。（剩余8366字）