Camera Movement Estimation and Path Correction using the Combination of Modified A-SIFT and Stereo System for 3D Modelling

Usha Kumari, Shuvendu Rana · 2025

Read Camera Movement Estimation and Path Correction using the Combination of Modified A-SIFT and Stereo System for 3D Modelling by Usha Kumari, Shuvendu Rana (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-17668.txt.

Introduction

Kumari and Rana improve how camera paths are tracked to build accurate three dimensional models in twenty twenty-five. Wide swings in view angle and drift in alignment make the task hard. Their tweak to a feature matching tool pulls more match points at lower cost. A two camera setup then fixes small errors in rotation.

A stereo camera model gauges how the camera shifts through space and corrects its motion. The full system matched the true camera path almost exactly in tests. It beat current path finding methods on accuracy. The authors pitch it for high fidelity three dimensional rebuilding in fields like mapping and film.

Catalog record

CREATORUsha Kumari, Shuvendu Rana
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-17668

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Camera Movement Estimation and Path Correction using the Combination of Modified A-SIFT and Stereo System for 3D Modelling Usha Kumari, Shuvendu Rana Abstract Creating accurate and efficient 3D models poses significant challenges, particularly in addressing large viewpoint variations, computational complexity, and alignment discrepancies. Efficient camera path generation can help resolve these issues. In this context, a modified version of the Affine Scale-Invariant Feature Transform (ASIFT) is proposed to extract more matching points with reduced computational overhead, ensuring an adequate number of inliers for precise camera rotation angle estimation. Additionally, a novel two-camera-based rotation correction model is introduced to mitigate small rotational errors, further enhancing accuracy. Furthermore, a stereo camera-based translation estimation and correction model is implemented to determine camera movement in 3D space by altering the Structure From Motion (SFM) model. Finally, the novel combination of ASIFT and two camera-based SFM models provides an accurate camera movement trajectory in 3D space. Experimental results show that the proposed camera movement approach achieves 99.9% accuracy compared to the actual camera movement path and outperforms state-of-the-art camera path estimation methods. By leveraging this accurate camera path, the system facilitates the creation of precise 3D models, making it a robust solution for applications requiring high fidelity and efficiency in 3D reconstruction. 2024 keywords Simultaneous Localization and Mapping, Structure From Motion, multi-view, Affine SIFT 1 Introduction Technological transitions have been evolving over the past few years. One such revolution can be seen in the landscape of 3D modeling. The role of 3D modeling is crucial in various industries, as it visualizes a product or concept before it is physically made, saving both time and resources. In the context of this field, an accurate 3D modeling technique becomes even more important. Recently, cameras using Simultaneous Localization and Mapping (SLAM) [ 1 ] have been employed to create 3D structures. The main challenge in this process is creating and detecting the camera movement path. Due to the use of non-rigid cameras for 3D model creation, proper camera localization techniques have become an important research area. In this regard, Structure From Motion (SFM) [ 2 ] plays an important role in estimating the camera movement path. However, the primary issue is that estimating accurate world coordinate mapping using SFM becomes difficult due to the non-rigid movement of the camera. Moreover, the presence of incorrect inliers in the SFM model leads to errors in camera position estimation. In 3D modeling technique, many works have been carried out [ 2 , 3 , 1 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 ] . Some of the works are based on 3D surface estimation, Camera positioning and Non-rigid object localization [ 29 ] . In the realm of three-dimensional reconstruction, accurate surface estimation is crucial. This becomes even more important when dealing with objects that have complex surface characteristics, such as specular reflectance. In this area, Jiang Yu Zheng et al. [ 2 ] propose a 3D reconstruction method that utilizes object rotation and extended lighting. This method is based on linear equations for camera motion and surface estimation. However, the approach is sensitive to changes in lighting and faces challenges with shadows, rough surfaces, and complex geometries. In order to determine the best camera placement, Robert Bodor et al. [ 26 ] present an optimization strategy that balances coverage and resolution by adjusting camera settings iteratively based on motion statistics. The method is effective but can be computationally expensive, particularly in complex multi-camera setups. Ware and Osborne [ 27 ] highlighted the difficulty of using six-degree-of-freedom (DOF) methods for 3D camera control with 2D inputs, which led Hachet et al. [ 28 ] to develop the Navidget approach for simplified control. Navidget is more effective and intuitive, especially with its circular metaphor for camera movement. However, it may struggle in crowded situations with frequent occlusions and requires more processing power. Using the pixel-wise eigenspace method, Arif and Vela [ 29 ] propose a method for non-rigid object localization and segmentation, enhancing deformation resilience through mean-shift iterations and similarity measures, despite facing challenges with significant appearance changes and severe occlusions. Similarly, Comaniciu et al. [ 30 ] present a non-rigid object localization technique based on color models and the mean-shift algorithm, yet face difficulties in dynamic scenes with occlusion or dynamic environments. On the other hand, in the case of SLAM-based techniques, Lu and Milios [ 4 ] presented a graph-based method for mapping using odometry and scan alignment, optimizing pose relations through scan matching. However, it relies heavily on accurate camera positioning systems. Furthermore, S. Thrun and M. Montemerlo [ 5 ] introduced Graph SLAM for offline SLAM by converting the SLAM posterior into a graphical network, using GPS data and variable elimination. However, it is limited by its assumption of independent Gaussian noise. Andrew J. Davison et al. [ 6 ] present a real-time algorithm for 3D trajectory recovery of a monocular camera, achieving drift-free SLAM performance. However, it lacks an in-depth discussion on real-world implementation challenges and sensor noise. Brian Williams et al. [ 7 ] developed a real-time monocular SLAM system using RANSAC and online learning. However, it struggles with relocalization beyond 80 features, complexity in large maps, and potential feature drift. Hauke Strasdat et al. [ 8 ] compare filtering and sparse bundle adjustment for visual SLAM using Monte Carlo experiments, emphasizing accuracy and cost. However, they overlook large-scale SLAM, loop closing, and the impact of the visual front end. Similarly, in this field, Raúl Mur-Artal et al. [ 9 ] introduce a real-time monocular SLAM system using ORB features for robust tracking and mapping. However, it can be complex for new users, resource-intensive, and has limited testing in extreme conditions. Erik Sandström [ 10 ] proposes a dense visual SLAM method with an adaptive point-based neural scene representation for improved accuracy and efficiency. However, it faces challenges related to complexity, resource demands, and generalizability. Fayad et al. [ 12 ] propose a quadratic deformation model for non-rigid SFM, enhancing 3D reconstruction of complex deformations. However, it requires precise initial estimates and careful parameter tuning. Paulo F. U. Gotardo and Aleix M. Martinez [ 13 ] presented a novel Non-Rigid Structure-from-Motion technique using complementary rank-3 spaces for improved shape reconstruction. However, it faces challenges with complex shapes, occlusions, resource demands, and generalizability. Suryansh Kumar et al. [ 14 ] present a method for multi-body non-rigid SFM, enabling simultaneous segmentation and reconstruction of non-rigid structures. However, it faces challenges related to computational complexity, parameter sensitivity, and limited generalization. Agniva Sengupta and Adrien Bartoli [ 15 ] introduce a tubular NRSfM method for improved 3D reconstruction in colonoscopy, achieving a 71.74% improvement. However, it faces challenges with weak textures and unstable geometry. M. J. Westoby et al. [ 16 ] highlight the cost-effectiveness of SFM photogrammetry for high-resolution topographic reconstruction in geoscience. However, they note limitations in error analysis, processing times, and performance in densely vegetated areas. Martínez-Espejo Zaragoza et al.

Read the complete text