现代信息科技2025，Vol.9Issue(22)：23-29,7.DOI:10.19850/j.cnki.2096-4706.2025.22.005
电梯检验视频的缓冲器行为AI跟踪判定方法研究
Research on AI Tracking and Judgment Method for Buffer Behavior in Elevator Inspection Videos
摘要
Abstract
The number of elevators in service is huge,and their periodic inspections form a massive video database.Manual review of video compliance has problems of low efficiency and low reliability.To address this,this paper studies a buffer behavior tracking and judgment method based on Mask R-CNN+SORT+HeightSeqAnalyzer.It uses Mask R-CNN and SORT algorithms respectively to achieve high-precision detection of buffer objects in single frames and object tracking in the temporal dimension.It calculates the reference height of objects through the pixel height sequence of the HeightSeqAnalyzer algorithm,judges object compression behavior according to the compression degree threshold and compression duration threshold,and finally merges compression events based on the overlap of compression frame sets to output the number of buffer compression events in the video.A training set and a validation set are constructed through orthogonal experimental design to train the Object Detection model.Experiments show that in image-level detection,the method in this paper achieves a precision P=0.81,recall R=0.78,mAP@0.75=0.72,and mAP@0.5:0.95=0.63 when both the IoU threshold and confidence threshold are 0.75.In video-level judgment,the precision P and specificity Sp both reach 1.00,and the accuracy A=0.84,completely eliminating FP samples,which verifies the effectiveness and engineering applicability of the method in this paper for automatic review of elevator inspection video compliance.关键词
行为判定/目标检测/目标跟踪/电梯检验Key words
behavior determination/Object Detection/object tracking/elevator inspection分类
信息技术与安全科学引用本文复制引用
罗伟立,陈贵龙,刘桂雄,陈建勋..电梯检验视频的缓冲器行为AI跟踪判定方法研究[J].现代信息科技,2025,9(22):23-29,7.基金项目
广东省市场监督管理局项目(2025CT03) （2025CT03）
