Research Of Detection System For Gun Barrel Damage Based On Active Stereo Omni-directional Vision Sensor

Gun dominates the land warfare,so countries around the world attach great importance to gun basic research.In the firing process,shells and propellant gas will cause physical or chemical damage to the barrel,there will be various kinds of damages such as cracks and corrosion,which seriously affect the firing accuracy,barrel life and shooting security.Therefore,how to quickly detect the damage of gun barrel has become an urgent problem to be solved.In order to address the difficulties in image deformation,seamless stitching and morphology data acquisition of the inner surface of small-bore gun barrel,this thesis design and realize a detection system for gun barrel damage based on active stereo omni-directional vision sensor,which collect panoramic images to detect the appearance of the barrel,collect laser scanning panoramic images to detect the shape of the barrel.Then fuse information to implement quantitative and qualitative analysis,which achieve the automatic vision detection process of “acquisition-recognition-judgment-reconstruction” for gun barrel detection.This thesis mainly focused on the following aspects:(1)Research on damage detection equipment of gun barrel.This thesis design and realize the damage detection equipment for gun barrel,the hardware mainly includes active panoramic vision sensor(active stereo omni-directional vision sensor,ASODVS),wireless communication unit,centering and moving device,LED lighting source,etc.The detection process is as follows: the staff promoted telescopic rod which drive the centering and moving device,the ASODVS fixed on the device collected the panoramic images of the inner wall of barrel in real time,and images were transmitted by wireless transmission units to PC and detected by the software.(2)Research on the damage recognition method based on convolutional neural network.Rifling is widely distributed throughout the barrel with complex environment,where has a large difference in the class and a small difference between the classes.We mainly use the following process to extract and recognize the gun barrel damage: Firstly,the panoramic images were preprocessed by the panoramic image unwrapped,light intensity adjusted and rifling eliminated,etc.;Secondly,the damages of binary images by optimum threshold value were extracted through connected regions;Finally,the damages were classified automatically under convolutional neural network model which was trained by damage data.(3)Research on the shape information detection method based on ASODVS.For the problem of multiple bore damage types,some damages can’t be recognized by appearance unless fusing the shape information.Therefore,this thesis use the following process to quantitative analysis including depth,area,deformation,etc.;Firstly,the laser light central points were extracted from laser panoramic images and smoothed;Secondly,spatial coordinates which reflect the shape of the inner gun barrel were calculated with the calibration result;Finally,the laser point cloud data was processed to damage information,and was reconstructed the true shape of barrel,which gives the inspectors an intuitive feel.