Gijs Dubbelman

Efficient trajectory bending with applications to loop closure

In robotic applications the absolute pose is often obtained as the integral of successive relative rigid-body motions. As each relative rigid-body motion is typically the product of statistical inference, the integrated absolute pose will exhibit error build-up and the estimated trajectory will differ from the true trajectory undertaken by the system. Some application areas allow the system to receive additional information about its current absolute pose, for example from loop detection, which is more accurate than the integral of the relative rigid-body motions. The availability of this absolute information is usually less frequent than the information underlying the relative rigid-body motions. This contribution addresses an efficient closed form algorithm which minimally bends a trajectory such that the integrated pose is exactly equal to any particular desired pose. The manner in which the bending is distributed over the trajectory is controllable using weights. The proposed method will be compared against a maximum likelihood solution on simulated trajectories as well as on trajectories estimated from binocular and monocular data. The results indicate that the performance differences between the closed form approach and the maximum likelihood solution are negligible while the closed form approach is significantly more efficient.

Closed-form Online Pose-chain SLAM

A novel closed-form solution for pose-graph SLAM is presented. It optimizes pose-graphs of particular structure called pose-chains by employing an extended version of trajectory bending. Our solution is designed as a back-end optimizer to be used within systems whose front-end performs state-of-the-art visual odometry and appearance based loop detection. The optimality conditions of our closed-form method and that of state-of-the-art iterative methods are discussed. The practical relevance of their theoretical differences is investigated by extensive experiments using simulated and real data. It is shown using 49 kilometers of challenging binocular data that the accuracy obtained by our closed-form solution is comparable to that of state-of-the-art iterative solutions while the time it needs to compute its solution is a factor 50 to 200 times lower. This makes our approach relevant to a broad range of applications and computational platforms.

Extending the Stixel World with online self-supervised color modeling for road-versus-obstacle segmentation

This work concentrates on vision processing for ADAS and intelligent vehicle applications. We propose a color extension to the disparity-based Stixel World method, so that the road can be robustly distinguished from obstacles with respect to erroneous disparity measurements. Our extension learns color appearance models for road and obstacle classes in an online and self-supervised fashion. The algorithm is tightly integrated within the core of the optimization process of the original Stixel World, allowing for strong fusion of the disparity and color signals. We perform an extensive evaluation, including different self-supervised learning strategies and different color models. Our newly recorded, publicly available data set is intentionally focused on challenging traffic scenes with many low-texture regions, causing numerous disparity artifacts. In this evaluation, we increase the F-score of the drivable distance from 0.86 to 0.97, compared to a tuned version of the state-of-the-art baseline method. This clearly shows that our color extension increases the robustness of the Stixel World, by reducing the number of falsely detected obstacles while not deteriorating the detection of true obstacles.

COP-SLAM:Closed-form online pose-chain optimization for visual SLAM

In this paper, we analyze and extend the recently proposed closed-form online pose-chain simultaneous localization and mapping (SLAM) algorithm. Pose-chains are a specific type of extremely sparse pose-graphs and a product of contemporary SLAM front-ends, which perform accurate visual odometry and reliable appearance-based loop detection. They are relevant for challenging robotic applications in large-scale 3-D environments for which frequent loop detection is not desired or not possible. Closed-form online pose-chain SLAM efficiently and accurately optimizes pose-chains by exploiting their Lie group structure. The convergence and optimality properties of this solution are discussed in detail and are compared against state-of-the-art iterative methods. We also provide a novel solution space, that of similarity transforms, which has not been considered earlier for the proposed algorithm. This allows for closed-form optimization of pose-chains that exhibit scale drift, which is important to monocular SLAM systems. On the basis of extensive experiments, specifically targeting 3-D pose-chains and using a total of 60 km of challenging binocular and monocular data, it is shown that the accuracy obtained by closed-form online pose-chain SLAM is comparable with that of state-of-the-art iterative methods, while the time it needs to compute its solution is orders of magnitudes lower. This novel SLAM technique thereby is relevant to a broad range of robotic applications and computational platforms.

Free-space detection with self-supervised and online trained fully convolutional networks

Recently, vision-based Advanced Driver Assist Systems have gained broad interest. In this work, we investigate free-space detection, for which we propose to employ a Fully Convolutional Network (FCN). We show that this FCN can be trained in a self-supervised manner and achieve similar results compared to training on manually annotated data, thereby reducing the need for large manually annotated training sets. To this end, our self-supervised training relies on a stereo-vision disparity system, to automatically generate (weak) training labels for the color-based FCN. Additionally, our self-supervised training facilitates online training of the FCN instead of offline. Consequently, given that the applied FCN is relatively small, the free-space analysis becomes highly adaptive to any traffic scene that the vehicle encounters. We have validated our algorithm using publicly available data and on a new challenging benchmark dataset that is released with this paper. Experiments show that the online training boosts performance with 5% when compared to offline training, both for Fmax and AP.

Color-Based Free-Space Segmentation Using Online Disparity-Supervised Learning

This work contributes to vision processing for Advanced Driver Assist Systems (ADAS) and intelligent vehicle applications. We propose a color-only stixel segmentation framework to segment traffic scenes into free, drivable space and obstacles, which has a reduced latency to improve the real-time processing capabilities. Our system learns color appearance models for free-space and obstacle classes in an online and self-supervised fashion. To this end, it applies a disparity-based segmentation, which can run in the background of the critical system path, either with a time delay of several frames or at a frame rate that is only a third of that of the color-based algorithm. In parallel, the most recent video frame is analyzed solely with these learned color appearance models, without an actual disparity estimate and the corresponding latency. This translates into a reduced response time from data acquisition to data analysis, which is a critical property for high-speed ADAS. Our evaluation on two publicly available datasets, one of which we introduce as part of this work, shows that the color-only analysis can achieve similar or even better results in difficult imaging conditions, compared to the disparity-only method. Our system improves the quality of the free-space analysis, while simultaneously lowering the latency and the computational load.

Environment mapping and localization with an uncontrolled swarm of ultrasound sensor motes

A method is presented in which a (large) swarm of sensor motes perform simple ultrasonic ranging measurements. The method allows to localize the motes within the swarm, and at the same time, map the environment which the swarm has traversed. The motes float passively uncontrolled through the environment and do not need any other sensor information or external reference other than a start and end point. Once the motes are retrieved, the stored data can be converted into the motes relative positions and a map describing the geometry of the environment. This method provides the possibility to map inaccessible or unknown environments where electro-magnetic signals, such as GPS or radio, cannot be used and where placing beacon points is very hard. An example is underground piping systems transporting liquids. Size and energy constraints together with the occurrence of reverberations pose challenges in the way the motes perform their measurements and collect their data. A minimalistic approach in the use of ultrasound is pursued, using an orthogonal frequency division multiplexing technique for the identification of motes. Simulations and scaled air-coupled 45–65 kHz experimental measurements have been performed and show feasibility of the concept.

Robust sensor cloud localization from range measurements

This work provides a feasibility study on estimating the 3-D locations of several thousand miniaturized free-floating sensor platforms. The localization is performed on basis of sparse ultrasound range measurements between sensor platforms and without the use of beacons. We show that this task can be viewed as a specific type of pose graph optimization. The main challenge is robustly estimating an initial pose graph, that models the locations of sensor platforms. For this, we introduce a novel graph growing strategy that uses random sample consensus in alternation with non-linear refinement. The theoretical properties of our sensor cloud localization method are analyzed and its robustness is investigated using simulations. These simulations are based on inlier-outlier measurement models and focus on the application of subterranean 3-D mapping of liquid environments, such as pipe infrastructures and oil wells.

Mapping swarms of resource-limited sensor motes: Solely using distance measurements and non-unique identifiers

This work is on 3-D localization of sensor motes in massive swarms based solely on 1-D relative distance-measurements between neighbouring motes. We target applications in remote and difficult-to-access environments such as the exploration and mapping of the interior of oil reservoirs where hundreds or thousands of motes are used. These applications bring forward the need to use highly miniaturized sensor motes of less than 1 centimeter, thereby significantly limiting measurement and processing capabilities. These constraints, in combination with additional limitations posed by the environments, impede the communication of unique hardware identifiers, as well as communication with external, fixed beacons.

On improving IED object detection by exploiting scene geometry using stereo processing

Detecting changes in the environment with respect to an earlier data acquisition is important for several applications, such as finding Improvised Explosive Devices (IEDs). We explore and evaluate the benefit of depth sensing in the context of automatic change detection, where an existing monocular system is extended with a second camera in a fixed stereo setup. We then propose an alternative frame registration that exploits scene geometry, in particular the ground plane. Furthermore, change characterization is applied to localized depth maps to distinguish between 3D physical changes and shadows, which solves one of the main challenges of a monocular system. The proposed system is evaluated on real-world acquisitions, containing geo-tagged test objects of 18 18 9 cm up to a distance of 60 meters. The proposed extensions lead to a significant reduction of the false-alarm rate by a factor of 3, while simultaneously improving the detection score with 5%.