Railway crossing abnormal event detection method and system based on generative adversarial network

Abstract

The invention discloses a railway crossing abnormal event detection method and system based on a generative adversarial network. The method comprises the following steps: acquiring a to-be-detected railway crossing video; processing a to-be-detected railway intersection video, judging whether abnormal event detection is needed, pausing abnormal event detection if a railway intersection handrail is lifted, and entering the next step if the railway intersection handrail is put down; and based on the trained generative adversarial network, performing abnormal event detection on the to-be-detected railway intersection video, and outputting an abnormal event detection result. An abnormity detection network architecture is provided based on the thought of a generative adversarial network, the appearance and motion features in a normal video are learned, and abnormity detection under the condition that negative samples are few or even no negative samples exist is achieved. A double-flow architecture is used, and a memory enhancement module is added, so that the capability of acquiring motion information by the network is improved, the capability of reconstructing the abnormal event by the network is inhibited, and the reconstruction error is increased, thereby further improving the detection effect of the abnormal event by the network.

Description

technical field [0001] The invention relates to the technical field of railway safety detection, in particular to a method and system for detecting abnormal events at railway crossings based on generative confrontation networks. Background technique [0002] The statements in this section merely mention the background technology related to the present invention and do not necessarily constitute the prior art. [0003] The purpose of abnormal event detection is to detect uncommon, abnormal objects or behaviors in videos or images that may pose hidden dangers to production and operation safety. [0004] The abnormal event detection of railway crossing is to monitor the railway crossing in real time, detect and locate the abnormal events in the monitoring area, such as: pedestrians stranded, landslides and falling rocks, etc. Through the real-time detection of the monitoring area, the factors that affect the safe operation of the train can be effectively discovered, and accura...

AI Technical Summary

Problems solved by technology

Most of the unguarded railway crossings just automatically drop the railings, but they cannot identify whether there are foreign objects or pedestrians stranded in the railway crossings.

Therefore, at present, most unmanned railway crossings still need a lot of manpower to be on duty before monitoring. Considering that the on-duty personnel usually cannot concentrate for a long time, it is difficult to achieve the desired effect

The data shows that there are still many railway crossings, which will consume a lot of manpower and material resources, and the safety is not high

[0006] At present, a series of anomaly detection methods for railway crossings have been proposed, but most of them are hardware-based methods with high cost and low intelligence

Common algorithms based on video images are greatly disturbed by factors such as light and shadow, and cannot guarantee the accuracy of detection in complex backgrounds, and there are problems such as waste of computing resources, and many types of abnormal events make it difficult to make data sets.

Images