| In the Big Data Era,single sensor alone is hard to obtain dataset with satisfying quality depicting the objects for different tasks.With the demand of overall understanding of entities in cities,combining data from multiple sensors become a hot topic[1].The IR inspection is usually done by analyzing IR images which is time consuming and inaccurate.Enriching the 3D MLS points with image sequences targets to combining thermal infrared images and the point cloud,and offer a 3D representation of the object with its temperature distribution/emission property,which helps with thermal building inspection,leakage detection,locating energy loss,and other tasks such as segmentation and classification.Although researchers presented different methods to combine the thermal infrared image with 3D data,seldom has been done on co-register IR images and Mobile laser scanner point clouds.Within the framework of this master thesis,the problem is to estimate the camera pose w.r.t.the point cloud.We proposed a semi-automatic process for single IR image and point cloud co-registration,which includes keypoints detection,corresponding point pairs generation,pose estimation,and 3D thermal model generation.An automatic method applying RANSAC was also tested.Besides,in order to achieve a low-cost process for image sequences,we tried to find the relative pose between images.A sfm method with semi-automatic co-registration was tested.In this thesis,we demonstrated that a 3D thermal model could be generated to enrich the information in MLS point cloud with IR images by co-registration.The result depends on the estimated camera pose.The semi-automatic method offering the best co-registration result.The result from the automatic camera pose estimation by RANSAC is limited due to the lack of correspondences between keypoints in images and point cloud.The smf can be used to reduce the computation by generating the relative pose between images,but the calibration parameters of IR camera needs to be accurate enough. |
PDF Full Text Request