Research On Multi-source Information Fusion Method For Night Vision Imaging System

The night vision system is an auxiliary device that provides safe and reliable information for vehicle driving at night.It can provide better vision and safer driving experience at night or under low light conditions.The vehicle night vision system can help drivers better deal with the challenges to night driving,such as blurred and incomplete vision.However,the current vehicle night vision system is composed of single sensor,which has problems such as poor robustness and lack of information.The multi-sensor system can solve these problems and make the imaging results more convenient for human eyes by integrating multi-mode graphics.Therefore,it is great significance to research the multi-source night vision imaging system.Aiming at a series of problems in the current night vision system,this paper focuses on the system design and multi-modal information fusion enhancement around the key technologies of multi-source night vision imaging system.The main research contents are as follows:1.In view of the night vision imaging scene,firstly,the night vision imaging system scheme is designed,and multiple sensors are selected and configured,mainly including visible camera,near infrared camera and LIDAR.Secondly,to solve the problem of different working voltages of multiple sensors,Buck-Boost voltage regulator module based on LTC3780 is designed to provide energy for multi-sensors.Finally,a multi-source night vision imaging system is built.2.According to the camera imaging model,the camera calibration process is be researched.The self-calibration of the camera is carried out by Zhang’s calibration method,and then the pose of the whole system is obtained through the joint calibration between the multi-modal camera and the lidar.To solve the problems of difficult registration and slow speed under night vision conditions,a registration method of visible light and near-infrared cameras based on lidar is proposed.The experimental results show that compared with the feature point registration algorithm,the algorithm has improved the operation efficiency by more than six times and has more advantages in image quality.3.This paper researches an edge-guided filtering algorithm.Firstly,color quantization is used to layer image colors,then the texture edge is extracted by texture edge detector.The core idea is to use pixel-neighborhood statistics to distinguish structure and texture.Secondly,under the guidance of the texture edge,the scale map is constructed,and then the input image is average filtered in the circular area of the scale map to get the coarse filtered image.Finally,a new pixel selection filter is used to optimize the filtering results.The proposed algorithm has higher indexes and achieves good results,and has achieved good results on SSIM and PSNR.It proves that the proposed algorithm can be used for image texture information extraction.4.A visible and near infrared image fusion algorithm based on texture information is studied to achieve night vision image enhancement.Firstly,the texture filter is used to separate the texture and structure of the multi-modal image.Secondly,since night vision images contain noise,Bayesian classification model is used to classify the features of texture images containing noise,then the noise is filtered by an improved joint filtering algorithm.Finally,the separated texture information and structure information are fused in the HSV color space.The algorithm in this paper has achieved higher indicators including VIFF,SSIM and Q^AB/Fin dataset.As for the field environment,the NIQE value of this algorithm is below 4.0,which has good results.In the case of night vision,not only the noise is eliminated,but also the details of the image are more complete,which is more conducive to human visual perception.In this paper,a night vision imaging system is built.In order to solve the problem of low efficiency in night vision images registration,a lidar-based registration method for visible light and near-infrared images is studied.Aiming at the problems of lack of details and noise interference in night vision images,a fusion enhancement algorithm of visible light and near-infrared images based on texture information is studied.Among them,to extract image structure and separate texture information,an edge-guided filtering algorithm is used.After experimental analysis and verification,the fusion enhancement algorithm in this paper has achieved good results and achieved the purpose of night vision image enhancement.The method is of great significance for improving the current vehicle night vision imaging level.