| Multi-sensor information fusion technology is focusing on how to use the information which is produced by many sensors in different places or different sensors in different places, to get a more reliable and more comprehensive description of a scene or a target reasonably. Now, Multi-sensor information fusion technology is a very important research content in image processing, pattern recognition and machine vision. The research achievement of Multi-sensor information fusion technology was applied to robot vision, military targets simulation, unmanned aerial vehicles navigation, driver assistance system and preservation of cultural.Many significant achievements and application cases have been achieved because of that a large study on multi-sensor information fusion technology been made by many experts and scholars at home and abroad these years. But there are still a lot of problems in the selection of sensors in different application fields, whether the information of one sensor is available in different moment and different sensors information fusion in multilayer and in multi-dimensions. This article fuses all these information of infrared, visible and laser range finder sensors to detect targets and then reconstruction it in natural scene. This article has researched on the extern calibration of a camera and a laser range finder, registration and fusion of visible image and infrared image, moving objects segmentation and 3-D reconstruction. Main research results include following:1. A cube with four face and five edges was designed for a laser range finder and visible sensor calibration. A connection between laser coordinate and image coordinate could be built by detecting three line equations of image through Hough transform and detecting laser point which is on the edge of the cube. After building the connection, solve the equations, then the rotation matrix and translation matrix could be solved.2. Combine the character of infrared image and visible image, a coarse to fine infrared image and visible image registration method has been proposed in this articlewhich is based on template image. Firstly get the template image of both sensors. Get the coarse registration parameters using Hough transform. Secondly, designed a new registration measurement called R-MI-Rényi which is combining Rényi mutual information and Harris corner function. Get the searching area and the size of registration window according to the coarse registration parameters. Finally get fine corresponding corners by using RANSAC, and then solve the fine registration parameters.3. The purpose of infrared image and visible image including on one hand reserve more detail and texture of visible image, on the hand highlight infrared object in infrared image, at the same time reduce distortion. A new image fusion method has been proposed in this article which is in NSCT domain based on PCNN and wavelet.Firstly decompose both images NSCT, then using PCNN to fuse the low frequency component and using wavelet to fuse the high frequency component.4. In order to solve the problem of object detection and 3-D reconstruction of it in natural scene. Firstly, detect the moving object in the fusion image of infrared and visible image by compute the optical flow. Secondly, reconstruct 3-D scene of moving object by the result of a camera and a laser range finder calibration. In the end fulfill target detection and reconstruction based on Multi-sensor information fusion。... |
PDF Full Text Request