Intelligent elevator door state detection method based on deep learning

Abstract

The invention discloses an intelligent elevator door state detection method based on deep learning. Through the analysis and recognition of a video monitoring image in an elevator car, the elevator door state is detected. Firstly, a target detection neural network model improved based on a yolov3-tiny framework is used for conducting target detection on an elevator door according to a certain period, and the location information of an elevator door target frame is detected, updated and saved. Every time when judging the elevator door state, aiming at a current elevator image, a picture is cutaccording to the saved location of the elevator target frame, then a user-defined efficient classification network model based on deep learning is used for classifying and analyzing the state of an elevator target frame picture, according to the three states of 'open', 'closed' and 'not closed well', a confidence probability value is given, and the state classification is conducted.

Description

technical field [0001] The invention relates to an intelligent detection method of an elevator door state based on deep learning technology, and belongs to the artificial intelligence monitoring and identification technology for elevator operation safety. Background technique [0002] Elevator is a vertical means of transportation widely and frequently used in modern human society. The safe operation of elevators is of paramount importance. However, elevators are complex mechanical equipment, and elevator trapping incidents caused by various reasons occur from time to time, and management and maintenance personnel cannot be automatically notified. At present, it is usually necessary to increase the way of the elevator Internet of Things to detect the occurrence of elevator trapped people in time. In this case, it is very important to detect the state judgment of the elevator door. Only by confirming that the elevator door has been closed and someone is active can the occur...

AI Technical Summary

Problems solved by technology

Generally speaking, neural network classification technology can be directly used to classify elevator car pictures, but the scenes in different elevator cars are changeable, and often due to the large number of passengers and severe occlusion, the direct classification effect is not good

Target detection technologies such as SSD and YOLO can also be directly used to identify targets in different door states. The target detection algorithm has high computational complexity, and generally has relatively high requirements for computing platforms, requiring GPUs as computing resources.

If real-time intelligent recognition is realized at the front end of the elevator, an embedded hardware platform such as artificial intelligence ASIC needs to be used, and the camera needs to be replaced or the intelligent recognition hardware needs to be added.

Images