Research On Vision Detection System Of Forest Grassland Inspection Robot

There are abundant natural resources hiding in the forest and grass environment,the protection and development for forest are of great significance to the ecological environment and human activities.With the development of robot technology and computer vision technology,it has become an important direction for the intelligent development of forest and grass equipment to use robots to replace manual inspection work and enter areas that are difficult for rescuers to reach in case of fire.Build a robot vision detection system,realize image information acquisition and use computer vision technology to complete the corresponding target intelligent detection has become a important mission in the intelligent forestry equipment.In the study,a forestry and grass inspection robot was designed around the ground inspection task of forestry and grass fires,a crawler type motion chassis structure was selected by comparing different robot motion structures,and a trinocular structure image acquisition device was designed around the construction of vision system,which is based on the theory of neural network and object detection to complete the construction of forest and grass fire identification network and conduct forest and grass fire detection;Based on binocular vision theory and stereo matching algorithm to study the localization of high temperature areas in forest and grass.The specific research contents are as follows:(1)Realize the hardware design of the vision system of the forest and grass inspection robot,mainly complete the selection of the robot motion structure,visible camera and thermal imager,and propose a trinocular vision structure.Combined with the characteristics of forest and grass environment,the crawler structure is determined as the motion structure of inspection robot,the different types of crawler motion chassis of different manufacturers are compared and analyzed,and the KOMODO-02 of Jichuang technology is determined as the final selection;Considering the visual field range,focal length,trigger mode and other factors of the visible light camera,Daheng MER-139-210U3 C is selected as the visible light acquisition hardware equipment of the inspection robot;The auxiliary detection is completed by using the FOTRIC-716 thermal imager and the visible light camera;Two visible light cameras and thermal imagers are used to build the visual inspection system of the inspection robot,and the mechanical structure supporting the trinocular vision system is designed;According to the hardware interface schematic diagram of camera and thermal imager,the trigger device of trinocular vision system is constructed by using Intel NUC8i5 BEK computing platform.(2)Collect data to bulid a forest and grass fire dataset named NEUF-Fire,and use YOLO v4-Tiny network for the fire detection,which has a pretty performance at the real-time detection.Labelimg is used for data annotation,and the NEFU-Fire forest and grass fire dataset used in the study is composed based on the pacsal VOC standard data set structure;YOLO v4-Tiny network is introduced into forest and grass fire detection.Aiming at the IOU calculation formula of clustering a priori frames using K-means algorithm in the network,the CIOU idea is introduced,and the distance calculation method in K-means is optimized by considering the factors such as the distance and aspect ratio between a priori frames.Analyze the relationship between IOU and K value,set the cluster as 4,set the learning rate subsection to speed up the convergence speed,compare and analyze the lost function,convergence speed,detection time and detection accuracy of YOLO V4,YOLO v4-Tiny and the algorithm proposed in the study,and prove the feasibility of the algorithm proposed in the study.(3)Study on the location of forest and grass high temperature area.Forest and grass fire detection can not provide the specific location of the fire.In order to provide guidance for the robot to quickly extinguish the fire,the positioning research is carried out by using binocular vision theory and combined with the image characteristics of thermal imager.Firstly,the parameters of binocular visible light system are calibrated and corrected;Secondly,due to the problems of occlusion in forest and grass environment,in order to make up for the defects of visible light camera,the thermal imager image and visible light image are combined in the research process,and the multi-source information image is obtained by using the principle of image linear mixing;SGBM stereo matching algorithm is used to obtain the target position and complete the positioning research.In conclusion,in order to protect the forest and grass resources and environment and improve the efficiency of forest and grass fire detection,the camera and thermal imager are used to build the hardware platform of the visual detection system of the forest and grass intelligent inspection robot,the YOLO v4-Tiny network is introduced to complete the fire detection,and the binocular vision and image linear mixing theory are used to complete the positioning research of the forest and grass high-temperature area.The experimental results show that the visual detection system of the forest and grass inspection robot can complete the monitoring task,find and locate the forest and grass fire in time,greatly improve the efficiency of forest and grass protection,and promote the forest and grass environmental protection and intelligent development of forest and grass.