Research On Distance Measurement Based On Infrared And Visible Binocular Vision

Binocular vision is a hot research field in current computer vision,which imitates human eyes to observe objects to obtain work scene information,locate target objects,and obtain depth information of target objects.Due to the complex and ever-changing surrounding environment and the increasing demand for information acquisition,conventional ranging methods such as ultrasound cannot meet the growing demand.However,depth information detection based on binocular vision can achieve functions such as depth detection,target recognition,and 3D information reconstruction,which has broad application prospects and value.In order to compensate for the shortcomings of visible light cameras as single spectral sensors in complex environments,such as heavy fog,strong light,or darkness,two infrared cameras can be used for binocular detection,or a binocular system combining an infrared camera and a visible light camera can be used for experiments and detection.The main research work of this article is as follows:(1)To use binocular ranging,it is first necessary to calibrate the binocular system,and then determine the coordinate conversion relationship of the target object based on the internal and external parameters of the binocular system obtained from the calibration.After comparing various commonly used calibration methods,the Zhang calibration method was used for calibration.Due to the inability of conventional paper calibration boards to meet the calibration requirements for infrared images,a specially made aluminum alloy checkerboard is used to calibrate both visible and infrared cameras.By heating the checkerboard,the contrast of the calibration board in infrared images was improved,and the calibration of infrared cameras was successfully achieved.The joint calibration of infrared cameras and visible light cameras was also achieved.The reprojection error of the dual visible binocular system is 0.06 pixel error,the reprojection error of the dual infrared binocular system is 0.14 pixel error,and the reprojection error of the visible and infrared binocular systems is 0.34 pixel error.All three binocular systems have achieved high precision joint calibration.(2)In the matching of dual visible light and dual infrared spectral images,the advantages and disadvantages of local stereo matching and global stereo matching were compared and analyzed.A semi global stereo matching algorithm combining these two methods was adopted and applied to the binocular system of dual visible light and the binocular ranging system of dual infrared thermal imaging spectral images.This semi global stereo matching algorithm has higher matching accuracy and faster matching speed compared to local stereo matching algorithms.The matching and ranging experimental results of the binocular system show that the target matching effect using semi global stereo matching based on path aggregation is good.The ranging error within 5 meters of the binocular ranging system with dual visible light is within 3%,and the ranging error within 5 meters of the binocular ranging system with dual infrared is within 5%.(3)In the matching of visible and infrared dual-spectral images,due to the different imaging mechanisms of infrared thermal imaging cameras and visible light cameras,the pixel grayscale values of infrared thermal imaging images are nonlinear,and the focal lengths,pixel sizes,and image sizes of infrared thermal imaging cameras and visible light cameras are different,making it difficult to extract common features of dual spectral images,It is not possible to use the same image feature stereo matching algorithm based on the size of image grayscale values as the same spectral image.In this paper,the method of combining phase information and mutual information is used to achieve stereo matching of dual-spectrum images.The root mean square error of the algorithm combining phase information and mutual information is 0.39,which is better than the root mean square error of 0.54 of the mutual information algorithm alone and 0.46 of the root mean square error of the algorithm combining mutual information and gradient information.The image matching accuracy is higher.The distance measurement error of the visible and infrared binocular system is within 5% within 3meters,and within 10% at 6 meters.