Bundle Adjustment (BA) takes a crucial part in Structure from Motion (SfM) which refines a visual reconstruction by optimizing the camera poses and feature positions. The performance of BA can differ depending on the parametrization methods. This paper evaluates two bundle adjustment techniques using standard BA function from MATLAB and Parallax BA. The two BA techniques are compared using data from the “Starry Night” and “MALAGA Parking-6L” with different initial inputs. The accuracy and convergence properties of the two BA methods have been evaluated. The effect of the different parameterization techniques and initial information was also analyzed. In most cases, the results of Parallax BA show better accuracy with lower final reprojection error and are less sensitive to the initialization values. It is evaluated that the parallax angle avoids the singularity issue commonly found in Standard BA, which shows that Parallax BA outperforms Standard BA. Furthermore, visual-inertial SLAM (VI-SLAM), based on Parallax BA, has been presented. It is much more reliable than a pure-vision system, showing further improved performance in terms of robustness and accuracy, even with less feature observation. The open-source code can be found in: https://github.com/uts-hb/ParallaxBA.git

We present a new multi-view 3D Euclidean reconstruction method for arbitrary uncalibrated radially-symmetric cameras, which needs no calibration or any camera model parameters other than radial symmetry. It is built on the radial 1D camera model [25], a unified mathematical abstraction to different types of radially-symmetric cameras. We formulate the problem of multi-view reconstruction for radial 1D cameras as a matrix rank minimization problem. Efficient implementation based on alternating direction continuation is proposed to handle scalability issue for real-world applications. Our method applies to a wide range of omni directional cameras including both dioptric and catadioptric (central and non-central) cameras. Additionally, our method deals with complete and incomplete measurements under a unified framework elegantly. Experiments on both synthetic and real images from various types of cameras validate the superior performance of our new method, in terms of numerical accuracy and robustness.

This paper describes an integrated navigation sensor module, including a camera, a laser scanner, and an inertial sensor, for unmanned aerial vehicles (UAVs) to fly both indoors and outdoors. The camera and the gimbaled laser sensor work in a complementary manner to extract feature points from the environment around the vehicle. The features are processed using an online extended Kalman filter (EKF) in simultaneous localization and mapping (SLAM) algorithm to estimate the navigational states of the vehicle. In this paper, a new method is proposed for calibrating a camera and a gimbaled laser sensor. This calibration method uses a simple visual marker to calibrate the camera and the laser scanner with each other. We also propose a real-time navigation algorithm based on the EKF SLAM algorithm, which is suitable for our camera-laser sensor package. The algorithm merges image features with laser range data for state estimation. Finally, these sensors and algorithms are implemented on our octo-rotor UAV platform and the result shows that our onboard navigation module can provide a real-time three-dimensional navigation solution without any assumptions or prior information on the surroundings.

This paper presents an efficient method of building continuous occupancy maps using Gaussian processes for large-scale environments. Although Gaussian processes have been successfully applied to map building, the applications are limited to small-scale environments due to the high computational complexity. To improve the scalability, we adopt a divide and conquer strategy where data are partitioned into manageable size of clusters and local Gaussian processes are applied to each cluster. Particularly, we propose overlapping clusters to mitigate the discontinuity problem that predictions of local estimators do not match along the boundaries. The results are consistent and continuous occupancy voxel maps in a fully Bayesian framework. We evaluate our method with simulated data and compare map accuracy and computational time with previous work. We also demonstrate our method with real data acquired from a laser range finder.

Although RGB-D sensors have been successfully applied to visual SLAM and surface reconstruction, most of the applications aim at visualization. In this paper, we propose a noble method of building continuous occupancy maps and reconstructing surfaces in a single framework for both navigation and visualization. Particularly, we apply a Bayesian nonparametric approach, Gaussian process classification, to occupancy mapping. However, it suffers from high-computational complexity of O(n(3))+O(n(2)m), where n and m are the numbers of training and test data, respectively, limiting its use for large-scale mapping with huge training data, which is common with high-resolution RGB-D sensors. Therefore, we partition both training and test data with a coarse-to-fine clustering method and apply Gaussian processes to each local clusters. In addition, we consider Gaussian processes as implicit functions, and thus extract iso-surfaces from the scalar fields, continuous occupancy maps, using marching cubes. By doing that, we are able to build two types of map representations within a single framework of Gaussian processes. Experimental results with 2-D simulated data show that the accuracy of our approximated method is comparable to previous work, while the computational time is dramatically reduced. We also demonstrate our method with 3-D real data to show its feasibility in large-scale environments.

This paper proposes a new method for occupancy map building using a mixture of Gaussian processes. We consider occupancy maps as a binary classification problem of positions being occupied or not, and apply Gaussian processes. Particularly, since the computational complexity of Gaussian processes grows as O(n3), where n is the number of data points, we divide the training data into small subsets and apply a mixture of Gaussian processes. The procedure of our map building method consists of three steps. First, we cluster acquired data by grouping laser hit points on the same line into the same cluster. Then, we build local occupancy maps by using Gaussian processes with clustered data. Finally, local occupancy maps are merged into one by using a mixture of Gaussian processes. Simulation results will be compared with previous researches and provided demonstrating the benefits of the approach.

Inertial-SLAM has been actively studied as it can provide all-terrain navigational capability with full six degrees-of-freedom information to autonomous robots. With the recent availability of low-cost inertial and vision sensors, a light-weight and accurate mapping system could be achieved for many robotic tasks such as land/aerial explorations. The key challenge toward this is in the availability of reliable and constant aiding information to correct the inertial system which is intrinsically unstable. The existing approaches have been relying on feature-based maps, which require accurate depth-resolution process to correct the inertial units properly where the aiding rate is highly dependent on the map density. In this work we propose to directly integrate the visual odometry to the inertial system by fusing the scale ambiguous translation vectors as Visual Directional Constraints (VDC) on vehicle motion at high update rates, while the 3D map being still used to constrain the longitudinal drifts but in a relaxed way. In this way, the visual odometry information can be seamlessly fused to inertial system by resolving the scale ambiguity problem between inertial and monocular camera thus achieving a reliable and constant aiding. The proposed approach is evaluated on SLAM benchmark dataset and simulated environment, showing a more stable and consistent performance of monocular inertial-SLAM.

This paper considers the question of using a nonlinear complementary filter for attitude estimation of fixed-wing unmanned aerial vehicle (UAV) given only measurements from a low-cost inertial measurement unit. A nonlinear complementary filter is proposed that combines accelerometer output for low frequency attitude estimation with integrated gyrometer output for high frequency estimation. The raw accelerometer output includes a component corresponding to airframe acceleration, occurring primarily when the aircraft turns, as well as the gravitational acceleration that is required for the filter. The airframe acceleration is estimated using a simple centripetal force model (based on additional airspeed measurements), augmented by a first order dynamic model for angle-of-attack, and used to obtain estimates of the gravitational direction independent of the airplane manoeuvres. Experimental results are provided on a real-world data set and the performance of the filter is evaluated against the output from a full GPS/INS that was available for the data set.