Stereo vision for simultaneous localization and mapping

[摘要] ENGLISH ABSTRACT: Simultaneous localization and mapping (SLAM) is vital for autonomous robot navigation. The robotmust build a map of its environment while tracking its own motion through that map. Althoughmany solutions to this intricate problem have been proposed, one of the most prominent issues thatstill needs to be resolved is to accurately measure and track landmarks over time. In this thesis weinvestigate the use of stereo vision for this purpose.In order to find landmarks in images we explore the use of two feature detectors: the scale-invariantfeature transform (SIFT) and speeded-up robust features (SURF). Both these algorithms find salientpoints in images and calculate a descriptor for each point that is invariant to scale, rotation andillumination. By using the descriptors we match these image features between stereo images anduse the geometry of the system to calculate a set of 3D landmark measurements. A Taylor approximationof this transformation is used to derive a Gaussian noise model for the measurements.The measured landmarks are matched to landmarks in a map to find correspondences. We find thatthis process often incorrectly matches ambiguous landmarks. To find these mismatches we developa novel outlier detection scheme based on the random sample consensus (RANSAC) framework. Weuse a similarity transformation for the RANSAC model and derive a probabilistic consensus measurethat takes the uncertainties of landmark locations into account. Through simulation and practicaltests we find that this method is a significant improvement on the standard approach of using thefundamental matrix.With accurately identified landmarks we are able to perform SLAM. We investigate the use of threepopular SLAM algorithms: EKF SLAM, FastSLAM and FastSLAM 2. EKF SLAM uses a Gaussiandistribution to describe the systems states and linearizes the motion and measurement equationswith Taylor approximations. The two FastSLAM algorithms are based on the Rao-Blackwellizedparticle filter that uses particles to describe the robot states, and EKFs to estimate the landmarkstates. FastSLAM 2 uses a refinement process to decrease the size of the proposal distribution andin doing so decreases the number of particles needed for accurate SLAM.We test the three SLAM algorithms extensively in a simulation environment and find that all threeare capable of very accurate results under the right circumstances. EKF SLAM displays extremesensitivity to landmark mismatches. FastSLAM, on the other hand, is considerably more robustagainst landmark mismatches but is unable to describe the six-dimensional state vector required for3D SLAM. FastSLAM 2 offers a good compromise between efficiency and accuracy, and performswell overall.In order to evaluate the complete system we test it with real world data. We find that our outlierdetection algorithm is very effective and greatly increases the accuracy of the SLAM systems. Wecompare results obtained by all three SLAM systems, with both feature detection algorithms, againstDGPS ground truth data and achieve accuracies comparable to other state-of-the-art systems.From our results we conclude that stereo vision is viable as a sensor for SLAM.

[发布机构] Stellenbosch University