Intelligent Recognition of Smoking and Calling Behaviors for Safety Surveillance

Electronics, Journal Year: 2023, Volume and Issue: 12(15), P. 3225 - 3225

Published: July 26, 2023

Smoking and calling are two typical behaviors involved in public industrial safety that usually need to be strictly monitored even prohibited on many occasions. To resolve the problems of missed detection false existing traditional deep-learning-based behavior-recognition methods, an intelligent recognition method using a multi-task YOLOv4 (MT-YOLOv4) network combined with behavioral priors is proposed. The original taken as baseline improved proposed method. Firstly, K-means++ algorithm used re-cluster optimize anchor boxes, which set predefined bounding boxes capture scale aspect ratio specific objects. Then, divided into branches same blocks but independent tasks after shared feature extraction layer CSPDarknet-53, i.e., behavior-detection branch object-detection branch, predict their related objects respectively from input image or video frame. Finally, according preliminary predicted results branches, comprehensive reasoning rules established obtain final result. A dataset smoking constructed for training testing, experimental indicate has 6.2% improvement recall 2.4% F1 score at cost slight loss precision compared method; achieved best performance among methods. It can deployed security surveillance systems unsafe-behavior monitoring early-warning management practical scenarios.

Language: Английский

Surveillance system to detect dangerous objects for children using YOLOv3 algorithm

AIP conference proceedings, Journal Year: 2023, Volume and Issue: 2977, P. 020072 - 020072

Published: Jan. 1, 2023

Language: Английский