Research On Multi-exposure Fusion Imaging Method For Calcium Carbide Furnace Scene

China's calcium carbide production industry is huge and plays an important role in the development of the national economy.Calcium carbide furnace is the highest labor cost and the most dangerous factor in calcium carbide production.At present,a robot which can replace artificial calcium carbide furnace has been developed.During the operation of the robot,the operator manipulate the robot with the real-time image sent back.The imaging dynamic range of the scene is very high,far beyond the dynamic range of the ordinary camera.This leads to the returned real-time images can not simultaneously image the details of the high-temperature fused calcium carbide and the operating environment around the robot.Due to the loss of on-site information,the operation efficiency of calcium carbide robot is greatly reduced,and it also brings great risk.Aiming at the operation scenario of calcium carbide robot,this thesis studies the imaging method for high dynamic scene based on multi exposure image fusion.Firstly,according to the dynamic range characteristics of calcium carbide furnace,multi-exposure control method is studied to establish long and short exposure channels.The dynamic range characteristics of the calcium carbide furnace are analyzed.The field luminance is mainly distributed in the high luminance fused calcium carbide and the relatively low brightness robot operating environment,so the strategy of automatic exposure control for each of them is determined respectively.Then,by studying the principle of automatic exposure control,the exposure time is taken as the control amount and the photometric method based on image processing is determined.On this basis,a multi-exposure control method based on light and dark partitions is proposed.First make adaptive dark and light partitions on live image.A coarse exposure adjustment link was designed.The average gray value of the light and dark partitions was used as the exposure control amount,and the exposure parameters were quickly adjusted to a reasonable range.Then the exposure precision adjustment link was designed.The image edge information values of the light and dark partitions are used as the exposure control feedback amount,and the one-dimensional search method is used to find the respective optimal exposure times of the light and dark partitions.Secondly,aiming at the real-time requirements of multi-exposure fusion,the multiexposure fusion image effect is guaranteed while the computational complexity of the algorithm is reduced,so that real-time multi-exposure image fusion is facilitated.On the basis of improving the classic algorithm,a fast multi-resolution multi-exposure image fusion method in YUV space is proposed,which converts multi-exposure image fusion to YUV color space.The fusion efficiency is increased by 3 times as the fusion effect is slightly higher than the classical algorithm,because only the fusion of the Y component and the more simple method of establishing the weight map.Thirdly,based on the GPU to speed up the relevant image processing process.The GPU-side multi-exposure control and multi-exposure fusion are implemented using the CUDA design program.For multi-exposure control,it takes only 20 ms for an iterative process,and can achieve a processing speed of 50 frames per second,which is higher than the image acquisition frame rate of 30 frames per second.The multi-exposure fusion takes only 33 ms for one iteration to achieve a processing speed of 30.3 frames per second,which is higher than the real-time processing requirement of 25 frames per second.Finally,a software and hardware experimental platform is set up for the scene of calcium carbide furnace,and the experimental verification algorithm is designed.facing with light changes in the scene of calcium carbide furnace,exposure coarse adjustment can effectively and quickly adjust the exposure time to a reasonable exposure range,the longest adjustment cycle is generally less than 300 ms.The exposure precision adjustment link can effectively find the maximum exposure time of the edge information through the one-dimensional search method under the condition that the brightness change is not severe.The entire exposure fine-tuning adjustment time is 330 ms.Based on this,realtime multi-exposure fusion image real-time output of calcium carbide furnace scene is realized,and the actual average frame rate reaches 28 frames per second.The obtained real-time image is clearly imaged on the details of the fused calcium carbide and the robot's operating environment,and the image exposure range and the existing imaging dynamic range are significantly improved compared to the long and short exposure channel images.