A new feature parametrization for monocular SLAM using line features

Liang Zhao, Shoudong Huang, Lei Yan, Gamini Dissanayake
2014 Robotica (Cambridge. Print)  
This paper presents a new monocular SLAM algorithm that uses straight lines extracted from images to represent the environment. A line is parametrized by two pairs of azimuth and elevation angles together with the two corresponding camera centres as anchors making the feature initialization relatively straightforward. There is no redundancy in the state vector as this is a minimal representation. A bundle adjustment (BA) algorithm that minimizes the reprojection error of the line features is
more » ... line features is developed for solving the monocular SLAM problem with only line features. A new map joining algorithm which can automatically optimize the relative scales of the local maps is used to combine the local maps generated using BA. Results from both simulations and experimental datasets are used to demonstrate the accuracy and consistency of the proposed BA and map joining algorithms. SLAM is the SLAM problem where the only sensor onboard the robot for observing the environment is a single camera [1], which is more challenging than the SLAM problems using laser sensors and/or RGB-D cameras because of the lack of depth information from the sensor measurements. Point features are commonly used in monocular SLAM because they are relatively easy to extract, match and represent. However, straight lines are very common in structured environments and arguably provide a better representation of the environment. Line features are less sensitive to motion blur [2], and especially suitable for environments with special structure. Thus, monocular SLAM using straight lines to represent the environment serves as a valuable addition to the suite of SLAM algorithms using monocular cameras. Feature based SLAM, whether solved using an estimation framework such as the extended Kalman filter (EKF) or an optimization framework, for example bundle adjustment (BA), requires the features to be represented in a state vector using an appropriate parametrization. Most of the line feature parametrizations proposed in the literature, for example traditional Plücker and Plücker based representations, are redundant [3] . Thus it is essential that the relationship between these parameters is imposed as a constraint during the SLAM process. In general, constrained optimization problems are more difficult to be solved than unconstrained optimization problems especially for high dimensional problems. Although constraints can be imposed as a pseudomeasurement in an estimation framework, this can lead to significant numerical issues [4] . Furthermore, some recent research [5] has raised questions about the theoretical validity of the pseudo-measurement approach to imposing constraints in an EKF framework. Clearly, an appropriate minimal representation provides significant advantages in this context. Thus in this paper, we only focus on minimal parametrizations to present line features in 3D environment. Bartoli and Strum [6] provided an orthonormal representation of the Plücker coordinates using minimal 4 parameters to represent a 3D line feature. A line in the environment is represented as a 3 × 3 and a 2 × 2 orthonormal matrices corresponding to its Plücker coordinates, and the 4 parameters can be used to update the Plücker coordinates during BA. Because of using
doi:10.1017/s026357471400040x fatcat:p6nnrm2mcnhu5dmy6jyzswvxdi