A real-time flame detection method for inspection robots

Abstract

The invention proposes a real-time flame detection method for an inspection robot, which includes: collecting image frame data of a moving vehicle-mounted camera, preprocessing, evaluating the clarity of the image, and applying a motion blur restoration algorithm when conditions are not met; The trained deep convolutional network model extracts the depth features of the preprocessed images; constructs a flame detector based on the aforementioned depth features and support vector machines; uses the flame detector in the discrimination stage to perform object detection on the image. The flame image detection method proposed by the invention can effectively complete the visual flame detection problem, and improve the efficiency and safety of flame recognition. The invention uses the features of deep convolution network calculation to replace the traditional artificial experience features to extract the relevant information of the flame, has the advantages of high precision, high reliability and strong robustness, and is suitable for the application of the inspection robot for real-time flame monitoring tasks need.

Description

technical field [0001] The invention relates to the technical field of computer vision and pattern recognition, in particular to a real-time flame detection method for an inspection robot. Background technique [0002] With the continuous progress of industrialization, the continuous liberation of productivity, and the gradual improvement of living standards, building facilities and materials have become more and more dense and diversified, resulting in more and more fire factors. higher challenges. Fire is one of the most common and most common major disasters that threaten public safety and social development. Its occurrence is often accompanied by huge direct or indirect economic losses, and even threatens human life. According to statistics, in the first half of 2018 alone, there were more than 160,000 fire alarms in my country. In response to fire, ancient people have long concluded the experience of "prevention is the first, rescue is the second, and caution is the n...

AI Technical Summary

Problems solved by technology

Considering that the movement of automatic inspection robots increases the difficulty of fire danger detection, that is, the previous method of using motion information to extract flame areas with fixed cameras or using context information through background modeling cannot meet this specific requirement.

Images