Research On Binocular Recognitoin Fire Detection Technology

As one of the most common sudden disasters,fires bring great damage to the bodies,minds and properties of people who encounter misfortune,and cause significant losses to the country’s economic development.In order to minimize the damage caused by fire to humans and the environment,this paper proposes the study of binocular recognition of fire technology,which has major study and practical advantages for improving the recognition accuracy of fire detection,achieving fire monitoring of large areas and high spatial resolution,improving the efficiency of fire handling,and the significance of fire prevention and resilience work.This paper investigates the detection and ranging of flames and smoke using YOLOv5 target detection technology and binocular stereo vision.The main aspects of the study are as follows.First,the basic theoretical knowledge related to the target detection algorithm and binocular stereo vision modeled by YOLOv5 is introduced respectively,which provides the theoretical basis for the next study.Second,an improved fire detection model for YOLOv5 is presented to address the problems of limited computational resources and large computation of the target detection model.The network model of the objective detection algorithm of YOLOv5 is improved by using a depth-separable convolutional network instead of the traditional convolution in the feature extraction network part to achieve a lightweight fire detection model with 51.6% less parameters and 46.2%less floating point operations compared to the original network.It satisfies real-time detection and is more conducive to model piggybacking on mobile devices.The CBAM attention mechanism is also added to the network model to improve the detection of small targets.The experimental results show that compared with the original YOLOv5 network structure,improving the model’s attention to target features can better handle complex scenarios,optimize the problem of occurrence of missed and false detection,improve the accuracy and stability of detected objects,and further improve the performance and robustness of the network.Then,in binocular stereo vision,the process of establishing and interconverting the relevant coordinate systems and camera calibration is introduced.The Binocular camera is standardized using MATLAB toolbox and the relevant parameters of the camera are imported into Bouguet algorithm in Open CV for stereo correction.In the stereo matching method,the SIFT algorithm is chosen for character point pairing and modified based on this algorithm,and the RANSAC algorithm is used for filtering.Through the experiment,the SIFT algorithm combined with RANSAC can effectively filter the key points of mis-matching,the number of mis-matching is reduced,and the accuracy and robustness of matching are improved.Finally,the improved YOLOv5 target detection algorithm is fused with the binocular stereo vision ranging algorithm to detect flames and smoke while also calculating the distance.This paper finally shows that the improved algorithm has been improved in both detection and ranging by experimental comparison.The algorithm study of binocular recognition fire detection technology is realized.