Three-dimensional image reconstruction using bundle adjustment applied to multiple texel images

Abstract

The importance of creating 3D imagery is increasing and has many applications in the field of disaster response, digital elevation models, object recognition, and cultural heritage. Several methods have been proposed to register texel images, which consist of fused lidar and digital imagery. The previous methods were limited to registering up to two texel images or multiple texel swaths having only one strip of lidar data per swath. One area of focus still remains to register multiple texel images to create a 3D model. The process of creating true 3D images using multiple texel images is described. The texel camera fuses the 2D digital image and calibrated 3D lidar data to form a texel image. The images are then taken from several perspectives and registered. The advantage of using multiple full frame texel images over 3D-or 2D-only methods is that there will be better registration between images because of the overlapping 3D points as well as 2D texture used in the joint registration process. The individual position and rotation mapping to a common world coordinate frame is calculated for each image and optimized. The proposed methods incorporate bundle adjustment for jointly optimizing the registration of multiple images. Sparsity is exploited as there is a lack of interaction between parameters of different cameras. Examples of the 3D model are shown and analyzed for numerical accuracy

Dynamic visualization of three-dimensional images from multiple texel images created from fused ladar/digital imagery

“…12 is composed of seven different 2.5-D texel images, which were captured sequentially in the CAIL laboratory using a handheld texel camera, and registered using the method described by Khatiwada and Budge. 4 The concept of weighted reliability is showcased using the "best" and "blended" texture assignment methods. For clarity, Figs.…”

“…2,3 Researchers at CAIL have exploited this data fusion to register multiple texel images into a single 3-D model. 4,5 The transformation parameters computed during registration are used to match individual texel images, bringing them into a single reference frame (or common world coordinate system). The result is a 3-D model composed of independent 2.5-D texel images.…”

GPS-denied navigation using location estimation and texel image correction

“…The research for this paper was designed in correspondence with previous research designed at the Center for Advanced Imaging LiDAR (CAIL). [4][5][6] This previous work involves multi-sensor fusion and registration for Textured Digital Surface Models (TDSM) from an sUAV payload for digital terrain modeling. This payload, called a "texel" camera, contains a Light Detection and Ranging (LiDAR) unit, an electro-optical (EO) camera, and an Internal Navigation System (INS).…”

Image enhancement method for visual localization and mapping in complex light scenarios

“…Based on this information, the three-dimensional mapping was conducted using forward intersection theory to calculate the three-dimensional coordinates of key points. Subsequently, the bundle adjustment algorithm was utilized to decrease the accumulated error from the process of relative orientation and forward intersection, optimizing the image position and coordinates of three-dimensional points simultaneously and globally, reducing reprojection error and improving the precision visual localization and mapping [20].…”