A Semantic Segmentation and Edge Detection Model Establishment and Guardrail Abnormality Monitoring Method

Abstract

The invention provides a semantic segmentation and edge detection model establishment and guardrail abnormality monitoring method, and relates to the technical field of guardrail monitoring. The semantic segmentation and edge detection model establishment method of the present invention includes: obtaining monitoring images of different roads under different perspectives, constructing a monitoring road data set according to the monitoring images; setting a loss function; according to the monitoring road data set and the obtained The loss function trains the preset convolutional neural network to output a semantic segmentation map and an edge detection map, until the loss function meets the preset conditions, and establishes a semantic segmentation and edge detection model according to the trained convolutional neural network . According to the technical solution of the present invention, the semantic segmentation and edge detection model is established by training the network that is mutually fused and promoted by the two channels of semantic segmentation and edge detection, so as to improve the fineness of semantic segmentation and the accuracy of edge detection, so that it can be used according to the model Effective identification and monitoring of abnormal guardrails.

Description

technical field [0001] The invention relates to the technical field of guardrail monitoring, in particular to a semantic segmentation and edge detection model establishment and a guardrail anomaly monitoring method. Background technique [0002] The isolation guardrail in the middle of the road is an important traffic infrastructure to maintain normal traffic order. It can effectively remind drivers of the driving area and separate two-way driving roads. At the same time, on the guardrails on both sides of the road, the division of motor vehicles and separate motor vehicle lanes is changed to prevent non-motor vehicles and pedestrians from crossing the motor vehicle lanes at will. [0003] At present, most guardrail detection belongs to the field of automatic driving. Guardrail data is collected through on-board cameras combined with lidar, and the cost of radar recognition is high. The method of vehicle-mounted monitoring guardrail can't accomplish real-time monitoring. I...

AI Technical Summary

Problems solved by technology

[0003] At present, most of the guardrail detection belongs to the field of automatic driving. The guardrail data is collected through the vehicle camera combined with the lidar, and the cost of the radar recognition solution is high.

The method of vehicle monitoring guardrail cannot achieve real-time monitoring

In addition, the current method has no special ability to identify the situation where the guardrail is blocked, and the false alarm rate is high. However, the blocked guardrail is often caused by passing vehicles due to camera vision occlusion, and it is not a real abnormality.

Images