This letter introduces an efficient bundle adjustment (BA) method based on the sparse Broyden-Fletcher-Goldfarb-Shanno (sparse BFGS, sBFGS) solution, which efficiently estimates camera poses and 3-D points. The Levenberg-Marquardt (LM) solution, which is widely applied in BA problems, requires more linear equation solution and more iterations. A gain matrix calculated using both the Jacobian matrix and residual vector replaces the simple diagonal matrix of the LM solution, which results in a better estimation of the descent direction and step size when finding a path to a local minimum in a sparse BFGS solution. Four datasets were verified, and the results demonstrate that the proposed method requires fewer linear equation solutions to converge to a minimum compared with that of the LM-based BA method.

The cost aggregation strategy shows a crucialrole in learning-based stereo matching tasks, where 3D convolutional filters obtain state of the art but require intensive computation resources, while 2D operations need less GPU memory but are sensitive to domain shift. In this letter, we decouple the 4D cubic cost volume used by 3D convolutional filters into sequential cost maps along the direction of disparity instead of dealing with it at once by exploiting a recurrent cost aggregation strategy. Furthermore, a novel recurrent module, Stacked Recurrent Hourglass (SRH), is proposed to process each cost map. Our hourglass network is constructed based on Gated Recurrent Units (GRUs) and down/upsampling layers, which provides GRUs larger receptive fields. Then two hourglass networks are stacked together, while multi-scale information is processed by skip connections to enhance the performance of the pipeline in textureless areas. The proposed architecture is implemented in an end-to-end pipeline and evaluated on public datasets, which reduces GPU memory consumption by up to 56.1% compared with PSMNet using stacked hourglass 3D CNNs without the degradation of accuracy. Then, we further demonstrate the scalability of the proposed method on several high-resolution pairs, while previously learned approaches often fail due to the memory constraint. The code is released at https://github.com/hongzhidu/SRHNet.

Tracking drift issues resulting from the lack of global positioning constraints prevent visual simultaneous localization and mapping (SLAM) systems from providing accurate pose estimation services, especially in indoor environments. Similar to the function of the Global Navigation Satellite System (GNSS), a new indoor equipment called workshop Measurement Positioning System (wMPS) has been developed to provide millimeter-level global positioning information in the large-volume metrology industrial community but is rarely applied to SLAM solutions. In this paper, we introduce a novel wMPS-SLAM framework that combines 3 degree-of-freedom (DoF) wMPS signals into our monocular pose estimation algorithm, enabling drift-free global 6DoF pose estimation in indoor scenarios. Firstly, a system calibration method is proposed to establish the relationship between the camera and wMPS and fuse the sensor information. Secondly, a variant similarity transformation scheme is employed to jointly estimate the scale factor and extrinsic parameters of the camera frame w.r.t the world frame for estimating the global camera poses. Lastly, a pose graph optimization method that minimizes global position and relative pose errors is utilized to fuse Visual Odometry (VO) estimation and wMPS measurements. Evaluation experiments conducted on two datasets demonstrate that wMPS-SLAM outperforms the state-of-the-art monocular visual solution (ORB-SLAM3) in terms of accuracy. Specifically, the absolute translation error (ATE) of the proposed system is reduced from 28.58mm to 7.018mm in real-world experiments.

For mobile robots, the high-precision integrated calibration and structural robustness of multi-sensor systems are important prerequisites for ensuring healthy operations in the later stage. Currently, there is no well-established validation method for the calibration accuracy and structural robustness of multi-sensor systems, especially for dynamic traveling situations. This paper presents a novel validation method for the calibration accuracy and structural robustness of a multi-sensor mobile robot. The method employs a ground–object–air cooperation mechanism, termed the “ground surface simulation field (GSSF)—mobile robot -photoelectric transmitter station (PTS)”. Firstly, a static high-precision GSSF is established with the true north datum as a unified reference. Secondly, a rotatable synchronous tracking system (PTS) is assembled to conduct real-time pose measurements for a mobile vehicle. The relationship between each sensor and the vehicle body is utilized to measure the dynamic pose of each sensor. Finally, the calibration accuracy and structural robustness of the sensors are dynamically evaluated. In this context, epipolar line alignment is employed to assess the accuracy of the evaluation of relative orientation calibration of binocular cameras. Point cloud projection and superposition are utilized to realize the evaluation of absolute calibration accuracy and structural robustness of individual sensors, including the navigation camera (Navcam), hazard avoidance camera (Hazcam), multispectral camera, time-of-flight depth camera (TOF), and light detection and ranging (LiDAR), with respect to the vehicle body. The experimental results demonstrate that the proposed method offers a reliable means of dynamic validation for the testing phase of a mobile robot.

Bundle adjustment (BA) using parallax angle feature parameterization (ParallaxBA), has been recently reported to have better convergence and more accurate structure estimation in. However, ParallaxBA is a merely free-net BA model and is incapable of geolocating tie points in the global (geodetic) coordinate system. This study presents a novel BA method that incorporates Ground Control Points (GCPs) into ParallaxBA based on hybrid feature parameterization (ParallaxBA-GCP). Different from other BA methods based on feature parameterization using Euclidean xyz coordinates, the tie points in ParallaxBA are parameterized with three types of angles: azimuth, elevation and parallax angles. Therefore, the observation function of the tie points and that of GCPs were described with hybrid feature parameterization in the form of angles and xyz parameterization, respectively. The result of one real data set demonstrates that, when a poor initialization is given, ParallaxBA-GCP can converge within a very few iterations and rapidly triangulate high-quality tie points that satisfies the accuracy requirement of the aerial triangulation.