In this paper we present a dense visual odometry system for RGB-D cameras performing both photometric and geometric error minimisation to estimate the camera motion between frames. Contrary to most works in the literature, we parametrise the geometric error by the inverse depth instead of the depth, which translates into a better fit of the distribution of the geometric error to the used robust cost functions. To improve the accuracy we propose to use a keyframe switching strategy based on a visibility criteria between frames. For the comparison of our approach with state-of-the-art approaches we use the popular datasets from the TUM for RGB-D benchmarking as well as two synthetic datasets. Our approach shows to be competitive with state-of-theart methods in terms of drift in meters per second, even compared to methods performing loop closure too. When comparing to approaches performing pure odometry like ours, our method outperforms them in the majority of the tested datasets. Additionally we show that our approach is able to work in real time and we provide a qualitative evaluation on our own sequences showing a low drift in the 3D reconstructions. We have implemented this method within the scope of PCL (Point Cloud Library) as a branch of the code for large scale KinectFusion, where the original ICP system for odometry estimation has been completely substituted by our method. A PCL fork including the modified method is available for download .

Stairs are one of the most common structures present in human-made scenarios, but also one of the most dangerous for those with vision problems. In this work we propose a complete method to detect, locate and parametrise stairs with a wearable RGB-D camera. Our algorithm uses the depth data to determine if the horizontal planes in the scene are valid steps of a staircase judging their dimensions and relative positions. As a result we obtain a scaled model of the staircase with the spatial location and orientation with respect to the subject. The visual odometry is also estimated to continuously recover the current position and orientation of the user while moving. This enhances the system giving the ability to come back to previously detected features and providing location awareness of the user during the climb. Simultaneously, the detection of the staircase during the traversal is used to correct the drift of the visual odometry. A comparison of results of the stair detection with other state-of-the-art algorithms was performed using public dataset. Additional experiments have also been carried out, recording our own natural scenes with a chest-mounted RGB-D camera in indoor scenarios. The algorithm is robust enough to work in real-time and even under partial occlusions of the stair.

Abstract-In this paper we present a dense visual odometry system for RGB-D cameras performing both photometric and geometric error minimisation to estimate the camera motion between frames. Contrary to most works in the literature, we parametrise the geometric error by the inverse depth instead of the depth, which translates into a better fit of the distribution of the geometric error to the used robust cost functions. We also provide a unified evaluation under the same framework of different estimators and ways of computing the scale of the residuals which can be found spread along the related literature. For the comparison of our approach with stateof-the-art approaches we use the popular dataset from the TUM for RGB-D benchmarking. Our approach shows to be competitive with state-of-the-art methods in terms of drift in meters per second, even compared to methods performing loop closure too. When comparing to approaches performing pure odometry like ours, our method outperforms them in the majority of the tested datasets. Additionally we show that our approach is able to work in real time and we provide a qualitative evaluation on our own sequences showing a low drift in the 3D reconstructions.

Abstract-In the last years monocular SLAM has been widely used to obtain highly accurate maps and trajectory estimations of a moving camera. However, one of the issues of this approach is that, due to the impossibility of the depth being measured in a single image, global scale is not observable and scene and camera motion can only be recovered up to scale. This problem gets aggravated as we deal with larger scenes since it is more likely that scale drift arises between different map portions and their corresponding motion estimates. To compute the absolute scale we need to know some kind of dimension of the scene (e.g., actual size of an element of the scene, velocity of the camera or baseline between two frames) and somehow integrate it in the SLAM estimation. In this paper, we present a method to recover the scale of the scene using an omnidirectional camera mounted on a helmet. The high precision of visual SLAM allows the head vertical oscillation during walking to be perceived in the trajectory estimation. By performing a spectral analysis on the camera vertical displacement, we can measure the step frequency. We relate the step frequency to the speed of the camera by an empirical formula based on biomedical experiments on human walking. This speed measurement is integrated in a particle filter to estimate the current scale factor and the 3D motion estimation with its true scale. We evaluated our approach using image sequences acquired while a person walks. Our experiments show that the proposed approach is able to cope with scale drift.

Autonomous navigation and recognition of the environment are fundamental abilities for people extensively studied in computer vision and robotics fields. Expansion of low cost wearable sensing provides interesting opportunities for assistance systems that augment people navigation and recognition capabilities. This work presents our wearable omnidirectional vision system and a novel twophase localization approach running on it. It runs stateof-the-art real time visual odometry adapted to catadioptric images augmented with topological-semantic information. The presented approach benefits from using wearable sensors to improve visual odometry results with true scaled solution. The wide field of view of catadioptric vision system used makes features last longer in the field of view and allows more compact location representation which facilitates topological place recognition. Experiments in this paper show promising ego-localization results in realistic settings, providing good true scaled visual odometry estimation and recognition of indoor regions.

In this paper we present a full-scaled real-time monocular SLAM using only a wearable camera. Assuming that the person is walking, the perception of the head oscillatory motion in the initial visual odometry estimate allows for the computation of a dynamic scale factor for static windows of N camera poses. Improving on this method we introduce a consistency test to detect non-walking situations and propose a sliding window approach to reduce the delay in the update of the scaled trajectory. We evaluate our approach experimentally on a unscaled visual odometry estimate obtained with a wearable camera along a path of 886 m. The results show a significant improvement respect to the initial unscaled estimate with a mean relative error of 0.91% over the total trajectory length.

In robot odometry and SLAM applications the real trajectory is estimated incrementally. This produces an accumulation of errors which gives raise to a drift in the trajectory. When revisiting a previous position this drift becomes observable and thus it can be corrected by applying loop closing techniques. Ultimately a loop closing process leads to an optimisation problem where new constraints between poses obtained from loop detection are applied to the initial incremental estimate of the trajectory. Typically this optimisation is jointly applied on the position and orientation of each pose of the robot using the state-of-the-art pose graph optimisation scheme on the manifold of the rigid body motions.In this paper we propose to address the loop closure problem using only the positions and thus removing the orientations from the optimisation vector. The novelty in our approach is that, instead of treating trajectory as a set of poses, we look at it as a curve in its pure mathematical meaning. We define an observation function which computes the estimate of one constraint in a local reference frame using only the robot positions. Our proposed method is compared against state-of-the-art pose graph optimisation algorithms in 2 and 3 dimensions. The main advantages of our method are the elimination of the need of mixing the orientation and position in the optimisation and the savings in computational cost due to the reduction of the dimension of the optimisation vector.

One open problem in the fields of place recognition and mapping is to be able to recognise a revisited place when its appearance and layout have changed between visits. In this paper, we investigate this problem in the context of RGB-D mapping in indoor environments. We propose to segment the scene in juts (neighbourhood of 3D points with normals that stick out from the surroundings) and look at low-level features, like textureness or entropy of the normals. These could differentiate those zones of the scene that change or move along time from those that are likely to remain static. We also present a method which improves the matching between images of the same place taken at different times by pruning details basing on these features. We evaluate on a number of communal areas and also on some scenes captured 6 months apart. Experiments with our approach, show an increase up to 70% in inlier matching ratio at the cost of pruning only less than 20% of correct matches, without the need of performing geometric verification.