Study On Automatic Fire Extinguishing Behavior Of Water Cannon Fire-fighting Robot Based On Binocular Vision

Fire-fighting robot can replace fire fighting personnel to enter high-risk fire field for fire fighting and rescue tasks,which can greatly reduce casualties and improve fire fighting efficiency.This paper aims at the problem that the existing fire fighting robots have poor autonomous combat ability and are difficult to meet the actual demand for fire fighting,and provides a new method for automatic fire fighting behavior of fire fighting robots.In the complex and changeable fire field,it is hard to realize the autonomy fire extinguishing operation of robot,so it is particularly important to judge the location of the fire source,strike the target accurately and formulate the fire extinguishing strategy.Therefore,fire sensing system built on multi-sensor fusion is explored to realize the identification of flame and temperature in the fire scene.The fire source location system based on binocular vision is explored and the space location of the flare target is realized.The target recognition system of fire water cannon jet track based on machine vision is studied,and the accurate judgment of water drop point is realized.Finally,the automated fire extinguishing operation was completed.To study fire environment information perception,the studied firefighting robot is equipped with a thermal imaging dual-spectrum MINI head.Background segmentation and motion detection are performed by multi-threshold Otsu method and inter-frame difference method;based on the typical characteristics of flame color,a combination of RGB and HSV based color model criterion is used to identify the frame target using a combination of multi-feature criterion.The flame target sample set is established based on digital image processing technology and trained with the help of a convolutional neural network.The YOLO target recognition algorithm is used to predict the image content and realize the accurate judgment of fire source location,which greatly improves the robustness and parallelism of the fire sensing system.In order to study the trajectory of the water jet and the location of the drop point of the fire-fighting water cannon,the main influencing factors of the change of the trajectory of the jet are analyzed;a model of the trajectory of the jet is established;the coordinates of the drop point and the range of the jet are deduced by extracting the jet curve;the location of the drop point of the water is judged by image processing from the first-person perspective with the help of corner point detection and k-means clustering algorithm in machine learning to provide information support for the automatic fire-fighting behavior.Based on the principle of binocular vision,the spatial location information of the fire source and the water fall point is judged.Based on the identified flame target and water fall image,the coordinate transformation relationship of binocular vision camera is established,and the two-dimensional coordinate information of the obtained target is used to judge the relative position relationship between the fire source target and the water fall point,and the automatic fire extinguishing behavior is realized by water cannon swing and pitch control.