Visual odometry using only a monocular camera faces more algorithmic challenges than stereo odometry. We present a robust monocular visual odometry framework for automotive applications. An extended propagation-based tracking framework is proposed which yields highly accurate (unscaled) pose estimates. Scale is supplied by ground plane pose estimation employing street pixel labeling using a convolutional neural network (CNN). The proposed framework has been extensively tested on the KITTI dataset and achieves a higher rank than current published state-of-the-art monocular methods in the KITTI odometry benchmark. Unlike other VO/SLAM methods, this result is achieved without loop closing mechanism, without RANSAC and also without multiframe bundle adjustment. Thus, we challenge the common belief that robust systems can only be built using iterative robustification tools like RANSAC. (C) 2017 Published by Elsevier B.V.

This paper presents a method for detecting independently moving objects (IMOs) from a monocular camera mounted on a moving car. We use an existing state of the art monocular sparse visual odometry/SLAM framework, and specifically attack the notorious problem of identifying those IMOs which move parallel to the ego-car motion, that is, in an `epipolar-conformant way. IMO candidate patches are obtained from an existing CNN-based car instance detector. While crossing IMOs can be identified as such by epipolar consistency checks, IMOs that move parallel to the camera motion are much harder to detect as their epipolar conformity allows to misinterpret them as static objects in a wrong distance. We employ a CNN to provide an appearance-based depth estimate, and the ambiguity problem can be solved through depth verification. The obtained motion labels (IMO/static) are then propagated over time using the combination of motion cues and appearance-based information of the IMO candidate patches. We evaluate the performance of our method on the KITTI dataset.

The recognition of individual object instances in single monocular images is still an incompletely solved task. In this work, we propose a new approach for detecting and separating vehicles in the context of autonomous driving. Our method uses the fully convolutional network (FCN) for semantic labeling and for estimating the boundary of each vehicle. Even though a contour is in general a one pixel wide structure which cannot be directly learned by a CNN, our network addresses this by providing areas around the contours. Based on these areas, we separate the individual vehicle instances. In our experiments, we show on two challenging datasets (Cityscapes and KITTI) that we achieve state-of-the-art performance, despite the usage of a subsampling rate of two. Our approach even outperforms all recent works w.r.t. several rating scores.

Most iterative optimization algorithms for motion, depth estimation or scene reconstruction, both sparse and dense, rely on a coarse and reliable dense initialization to bootstrap their optimization procedure. This makes techniques important that allow to obtain a dense but still approximative representation of a desired 2D structure (e.g., depth maps, optical flow, disparity maps) from a very sparse measurement of this structure. The method presented here exploits the complete information given by the principal component analysis (PCA), the principal basis and its prior distribution. The method is able to determine a dense reconstruction even if only a very sparse measurement is available. When facing such situations, typically the number of principal components is further reduced which results in a loss of expressiveness of the basis. We overcome this problem and inject prior knowledge in a maximum a posteriori (MAP) approach. We test our approach on the KITTI and the Virtual KITTI dataset and focus on the interpolation of depth maps for driving scenes. The evaluation of the results shows good agreement to the ground truth and is clearly superior to the results of an interpolation by the nearest neighbor method which disregards statistical information.

One of the major steps in visual environment perception for automotive applications is to track keypoints and to subsequently estimate egomotion and environment structure from the trajectories of these keypoints. This paper presents a propagation based tracking method to obtain the 2D trajectories of keypoints from a sequence of images in a monocular camera setup. Instead of relying on the classical RANSAC to obtain accurate keypoint correspondences, we steer the search for keypoint matches by means of propagating the estimated 3D position of the keypoint into the next frame and verifying the photometric consistency. In this process, we continuously predict, estimate and refine the frame-to-frame relative pose which induces the epipolar relation. Experiments on the KITTI dataset as well as on the synthetic COnGRATS dataset show promising results on the estimated courses and accurate keypoint trajectories.

Correspondence relations between different views of the same scene can be learnt in an unsupervised manner. We address autonomous learning of arbitrary fixed spatial (point-to-point) mappings. Since any such transformation can be represented by a permutation matrix, the signal model is a linear one, whereas the proposed analysis method, mainly based on Canonical Correlation Analysis (CCA) is based on a generalized eigensystem problem, i.e., a nonlinear operation. The learnt transformation is represented implicitly in terms of pairs of learned basis vectors and does neither use nor require an analytic/parametric expression for the latent mapping. We show how the rank of the signal that is shared among views may be determined from canonical correlations and how the overlapping (=shared) dimensions among the views may be inferred.

Phase correlation is one of the classic methods for sparse motion or displacement estimation. It is renowned in the literature for high precision and insensitivity against illumination variations. We propose several important enhancements to the phase correlation (PhC) method which render it more robust against those situations where a motion measurement is not possible (low structure, too much noise, too different image content in the corresponding measurement windows). This allows the method to perform self-diagnosis in adverse situations. Furthermore, we extend the PhC method by a robust scheme for detecting and classifying the presence of multiple motions and estimating their uncertainties. Experimental results on the Middlebury Stereo Dataset and on the KITTI Optical Flow Dataset show the potential offered by the enhanced method in contrast to the PhC implementation of OpenCV.

The online-estimation of yaw, pitch, and roll of a moving vehicle is an important ingredient for systems which estimate egomotion, and 3D structure of the environment in a moving vehicle from video information. We present an approach to estimate these angular changes from monocular visual data, based on the fact that the motion of far distant points is not dependent on translation, but only on the current rotation of the camera. The presented approach does not require features (corners, edges,...) to be extracted. It allows to estimate in parallel also the illumination changes from frame to frame, and thus allows to largely stabilize the estimation of image correspondences and motion vectors, which are most often central entities needed for computating scene structure, distances, etc. The method is significantly less complex and much faster than a full egomotion computation from features, such as PTAM [6], but it can be used for providing motion priors and reduce search spaces for more complex methods which perform a complete analysis of egomotion and dynamic 3D structure of the scene in which a vehicle moves.

Tracking keypoints through a video sequence is a crucial first step in the processing chain of many visual SLAM approaches. This paper presents a robust initialization method to provide the initial match for a keypoint tracker, from the 1st frame where a keypoint is detected to the 2nd frame, that is: when no depth information is available. We deal explicitly with the case of long displacements. The starting position is obtained through an optimization that employs a distribution of motion priors based on pyramidal phase correlation, and epipolar geometry constraints. Experiments on the KITTI dataset demonstrate the significant impact of applying a motion prior to the matching. We provide detailed comparisons to the state-of-the-art methods.

Visual odometry is one of the most active topics in computer vision. The automotive industry is particularly interested in this field due to the appeal of achieving a high degree of accuracy with inexpensive sensors such as cameras. The best results on this task are currently achieved by systems based on a calibrated stereo camera rig, whereas monocular systems are generally lagging behind in terms of performance. We hypothesise that this is due to stereo visual odometry being an inherently easier problem, rather than than due to higher quality of the state of the art stereo based algorithms. Under this hypothesis, techniques developed for monocular visual odometry systems would be, in general, more refined and robust since they have to deal with an intrinsically more difficult problem. In this work we present a novel stereo visual odometry system for automotive applications based on advanced monocular techniques. We show that the generalization of these techniques to the stereo case result in a significant improvement of the robustness and accuracy of stereo based visual odometry. We support our claims by the system results on the well known KITTI benchmark, achieving the top rank for visual only systems∗ .

We discuss matching measures (scores and residuals) for comparing image patches under unknown affine photometric (=intensity) transformations. In contrast to existing methods, we derive a fully symmetric matching measure which reflects the fact that both copies of the signal are affected by measurement errors (noise), not only one. As it turns out, this evolves into an eigensystem problem; however a simple direct solution for all entities of interest can be given. We strongly advocate for constraining the estimated gain ratio and the estimated mean value offset to realistic ranges, thus preventing the matching scheme from locking into unrealistic correspondences.

The development of vehicles that perceive their environment, in particular those using computer vision, indispensably requires large databases of sensor recordings obtained from real cars driven in realistic traffic situations. These datasets should be time shaped for enabling synchronization of sensor data from different sources. Furthermore, full surround environment perception requires high frame rates of synchronized omnidirectional video data to prevent information loss at any speeds.

This paper describes an experimental setup and software environment for recording such synchronized multi-sensor data streams and storing them in a new open source format. The dataset consists of sequences recorded in various environments from a car equipped with an omnidirectional multi-camera, height sensors, an IMU, a velocity sensor, and a GPS. The software environment for reading these data sets will be provided to the public, together with a collection of long multi-sensor and multi-camera data streams stored in the developed format.

We propose and evaluate a versatile scheme for image pre-segmentation that generates a partition of the image into a selectable number of patches (’superpixels’), under the constraint of obtaining maximum homogeneity of the ’texture’ inside of each patch, and maximum accordance of the contours with both the image content as well as a Gibbs-Markov random field model. In contrast to current state-of-the art approaches to superpixel segmentation, ’homogeneity’ does not limit itself to smooth region-internal signals and high feature value similarity between neighboring pixels, but is applicable also to highly textured scenes. The energy functional that is to be maximized for this purpose has only a very small number of design parameters, depending on the particular statistical model used for the images.

The capability of the resulting partitions to deform according to the image content can be controlled by a single parameter. We show by means of an extensive comparative experimental evaluation that the compactness-controlled contour-relaxed superpixels method outperforms the state-of-the art superpixel algorithms with respect to boundary recall and undersegmentation error while being faster or on a par with respect to runtime.

We present a novel method for clustering data drawn from a union of arbitrary dimensional subspaces, called Discriminative Subspace Clustering (DiSC). DiSC solves the subspace clustering problem by using a quadratic classifier trained from unlabeled data (clustering by classification). We generate labels by exploiting the locality of points from the same subspace and a basic affinity criterion. A number of classifiers are then diversely trained from different partitions of the data, and their results are combined together in an ensemble, in order to obtain the final clustering result. We have tested our method with 4 challenging datasets and compared against 8 state-of-the-art methods from literature. Our results show that DiSC is a very strong performer in both accuracy and robustness, and also of low computational complexity.

This book constitutes the refereed proceedings of the 33rd Symposium of the German Association for Pattern Recognition, DAGM 2011, held in Frankfurt/Main, Germany, in August/September 2011. The 20 revised full papers and 22 revised poster papers were carefully reviewed and selected from 98 submissions. The papers are organized in topical sections on object recognition, adverse vision conditions challenge, shape and matching, segmentation and early vision, robot vision, machine learning, and motion. The volume also includes the young researcher's forum, a section where a carefully jury-selected ensemble of young researchers present their Master thesis work.