摘要
Abstract
The analysis of long surveillance video data constitutes an effective strategy for advancing urban intelligence.We adopt an integrated methodology that combines the YOLO(You Only Look Once)object detection model,the DeepSort(Deep Learning Simple Online and Realtime Tracking)algorithm,and the SlowFast action recognition framework to extract and analyze pedestrian behavioral pat-terns from extensive video datasets.A dual-channel design is employed,in which feature extraction is performed independently on each video frame.To enhance the interconnectivity among diverse feature elements and ensure the comprehensive retention of frame informa-tion,the YOLO target detection model using dynamic detection head,the attention mechanism is integrated into the SlowFast action recog-nition framework.In the process of behavior analysis and classification,the one-hot encoding scheme is applied,followed by a dimension-ality reduction using Principal Component Analysis(PCA).The experimental results substantiate the efficacy of this approach in process-ing and analyzing behavioral data derived from long-duration video recordings.关键词
YOLO目标检测模型/DeepSort跟踪算法/SlowFast行为识别/主成分分析Key words
YOLO object detection model/DeepSORT tracking algorithm/SlowFast action recognition/Principal component analy-sis(PCA)分类
信息技术与安全科学
