Vehicle Odometry Research Articles

An Automatic Key-Frame Selection Method for Monocular Visual Odometry of Ground Vehicle

In response to the rapid change of attitude in complex situations such as sharp turns and camera jitter, visual odometry/visual simultaneous localization and mapping (VO/VSLAM) often requires a high camera frame rate. However, the high frame rate of the camera poses a great challenge to real-time VO/VSLAM, especially when the vehicle travels along a relatively flat trajectory which requires higher real-time performance. Therefore, in this paper, we propose an automatic method for key-frame selection based on the motion state of the vehicle, which can reduce data redundancy and improve the real-time performance and robustness of VO/VSLAM. First, a pyramid-layered Kanade-Lucas-Tomasi (KLT) algorithm is used to track the feature points, and the five-point method and RANSAC are used to calculate the essential matrix. Second, two-step decomposition is used to calculate the change of inter-frame attitude, and then, key-frames are automatically selected based on the motion state of the vehicle predicted by attitude change within a short time interval. To evaluate the method, we conduct extensive experiments on real data and Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) dataset based on monocular visual odometry. Then, we made qualitative and quantitative evaluations from the comparison with the reference trajectory, relative error (RE), root mean square error (RMSE), and absolute trajectory error (ATE). The results indicate that our method can improve the real-time performance while ensuring accuracy, with the data redundancy reduced by about 40%-60%. In addition, the performance of our method is further verified by comparison with the current representative ORB-SLAM.

Improving Low-Cost Solutions for Path Mapping in Autonomous Vehicle

One of the main challenge in designing autonomous vehicle is developing the algorithms necessary for simultaneous localization and mapping (SLAM). While this process heavily depends on an expensive hardware called Light Detection and Ranging (LiDAR), there are cheaper alternatives that can be implemented in earlier stage of autonomous vehicle development. Inertial Measurement Unit (IMU), vehicle odometry, and Global Positioning System (GPS) can be used as a solution. The main problem is these cheaper alternatives relies heavily on the precision of the hardware used. In this project, we outline the mapping process using these cheaper solutions which can also be used to complement LiDAR-based SLAM. We show that using simple approaches such as static bias drift removal, high-pass filtering, signal downsampling and linear interpolation, we can increase the robustness and improve the accuracy and precision of the IMU and GPS used respectively. We manage to increase the precision of the GPS readings and reduce the drift of IMU on average from -17.105 deg/min to -0.1177 deg/min. We show the improvement achieved by our proposed method by mapping the road around Engine Square, Jalan Ilmu 1/1, Universiti Teknologi MARA, Shah Alam, Selangor, Malaysia.

Hybrid real-time stereo visual odometry for unmanned aerial vehicles

A visual localization approach for unmanned aerial vehicles (UAVs) based on the hybrid real-time stereo visual odometry (VO) is presented. The hybrid VO initializes the semidirect visual odometry (SVO) with depth obtained from the stereo camera, which runs with the monocular scheme at 30 Hz. Meanwhile, a feature-based stereo VO runs in a parallel thread to improve the reliability. Considering the robustness of the feature-based VO, we not only are able to estimate the pose of the UAV with the same rate of the incoming images but also to recover the pose of the UAV when the SVO fails. We demonstrate the accuracy and the reliability of the hybrid VO based on a public benchmark dataset and a dataset recorded on our experimental platform. In addition, the autonomous hovering experiment verifies that the estimated result is fast and accurate enough to control the position of the UAV.

A Universal Vacant Parking Slot Recognition System Using Sensors Mounted on Off-the-Shelf Vehicles.

An automatic parking system is an essential part of autonomous driving, and it starts by recognizing vacant parking spaces. This paper proposes a method that can recognize various types of parking slot markings in a variety of lighting conditions including daytime, nighttime, and underground. The proposed method can readily be commercialized since it uses only those sensors already mounted on off-the-shelf vehicles: an around-view monitor (AVM) system, ultrasonic sensors, and in-vehicle motion sensors. This method first detects separating lines by extracting parallel line pairs from AVM images. Parking slot candidates are generated by pairing separating lines based on the geometric constraints of the parking slot. These candidates are confirmed by recognizing their entrance positions using line and corner features and classifying their occupancies using ultrasonic sensors. For more reliable recognition, this method uses the separating lines and parking slots not only found in the current image but also found in previous images by tracking their positions using the in-vehicle motion-sensor-based vehicle odometry. The proposed method was quantitatively evaluated using a dataset obtained during the day, night, and underground, and it outperformed previous methods by showing a 95.24% recall and a 97.64% precision.

Closed‐loop one‐way‐travel‐time navigation using low‐grade odometry for autonomous underwater vehicles

AbstractThis paper extends the progress of single beacon one‐way‐travel‐time (OWTT) range measurements for constraining XY position for autonomous underwater vehicles (AUV). Traditional navigation algorithms have used OWTT measurements to constrain an inertial navigation system aided by a Doppler Velocity Log (DVL). These methodologies limit AUV applications to where DVL bottom‐lock is available as well as the necessity for expensive strap‐down sensors, such as the DVL. Thus, deep water, mid‐water column research has mostly been left untouched, and vehicles that need expensive strap‐down sensors restrict the possibility of using multiple AUVs to explore a certain area. This work presents a solution for accurate navigation and localization using a vehicle's odometry determined by its dynamic model velocity and constrained by OWTT range measurements from a topside source beacon as well as other AUVs operating in proximity. We present a comparison of two navigation algorithms: an Extended Kalman Filter (EKF) and a Particle Filter(PF). Both of these algorithms also incorporate a water velocity bias estimator that further enhances the navigation accuracy and localization. Closed‐loop online field results on local waters as well as a real‐time implementation of two days field trials operating in Monterey Bay, California during the Keck Institute for Space Studies oceanographic research project prove the accuracy of this methodology with a root mean square error on the order of tens of meters compared to GPS position over a distance traveled of multiple kilometers.

Optimal Configuration of a Downward-Facing Monocular Camera for Visual Odometry

Background/Objectives: An accurate estimation of a robot’s pose is critical for autonomous mobile vehicles. Monocular vision-based odometry is one of the robust methods used for estimating the relative motion of vehicles by using only a stream of images acquired from a camera mounted on the vehicle. Methods: The location of the camera and its configuration can affect the quality of the localization process and the maximum permissible vehicle driving speed. In this work, the elements that can affect the permissible vehicle driving speed were recognized and the equations for this allowable speed were derived. Moreover, the optimal location and configuration of a downward-facing monocular camera that ensures the success of Visual Odometry for car-like ground vehicles in low-textured environment are presented after an extensive analysis. Findings: The results show that the suggested optimal camera configuration increases the permissible vehicle driving speed. Furthermore, the percentage of correlation mismatching between image frames is decreased as the suggested camera location avoids the negative consequences of shadows and directional sunlight that can add noise to image frames and interrupt the estimation of image pixel displacement. Application/Improvements: The suggested camera configuration could be implemented in various commercial mobile robotic applications, which utilize visual odometry for improved accuracy, efficiency and cost effectiveness.

Point Cloud Mapping Measurements Using Kinect RGB-D Sensor and Kinect Fusion for Visual Odometry

RGB-D camera like Kinect make available RGB Images along with per-pixel depth information in real time. This paper uses the Kinect Fusion developed by Microsoft Research for the 3D reconstruction of the scene in real time using the MicroKinect Camera and applies it as an aid for Visual Odometry of a Robotic Vehicle where no external reference like GPS is available.

Simplified epipolar geometry for real-time monocular visual odometry on roads

Simplified epipolar geometry is proposed in this paper to accelerate monocular visual odometry for ground vehicles. The vehicles on roads or indoors exhibit planar motion locally, and such prior has been effectively utilized to speed up monocular visual odometry. However, we observed that the previous planar motion models frequently failed because their over-simplification did not accept small non-planar motion caused by abrupt bumps or camera vibration. In this paper, simplified motion models are relaxed and their corresponding algorithms for relative pose estimation are derived. Effectiveness of the proposed algorithms is demonstrated by two types of experiments: relative pose estimation with synthetic data, and monocular visual odometry with real image sequences. In the first experiment, the proposed approximated 5-point algorithm provided similar (sometimes better) accuracy to the original 5-point algorithm, but it spent almost 15 times less computing time. In the second experiment, we also observed that monocular visual odometry with our algorithm had almost 9 times faster outlier rejection than previous approaches.

Multispectral Stereo Odometry

In this paper, we investigate the problem of visual odometry for ground vehicles based on the simultaneous utilization of multispectral cameras. It encompasses a stereo rig composed of an optical (visible) and thermal sensors. The novelty resides in the localization of the cameras as a stereo setup rather than two monocular cameras of different spectrums. To the best of our knowledge, this is the first time such task is attempted. Log-Gabor wavelets at different orientations and scales are used to extract interest points from both images. These are then described using a combination of frequency and spatial information within the local neighborhood. Matches between the pairs of multimodal images are computed using the cosine similarity function based on the descriptors. Pyramidal Lucas–Kanade tracker is also introduced to tackle temporal feature matching within challenging sequences of the data sets. The vehicle egomotion is computed from the triangulated 3-D points corresponding to the matched features. A windowed version of bundle adjustment incorporating Gauss–Newton optimization is utilized for motion estimation. An outlier removal scheme is also included within the framework to deal with outliers. Multispectral data sets were generated and used as test bed. They correspond to real outdoor scenarios captured using our multimodal setup. Finally, detailed results validating the proposed strategy are illustrated.

Integrated Indoor Navigation System for Ground Vehicles With Automatic 3-D Alignment and Position Initialization

This paper introduces an autonomous integrated indoor navigation system for ground vehicles that fuses inertial sensors, light detection and ranging (LiDAR) sensors, received signal strength (RSS) observations in wireless local area networks (WLANs), odometry, and predefined occupancy floor maps. This paper proposes a solution for the problem of automatic self-alignment and position initialization indoors under the absence of an absolute navigation system such as Global Navigation Satellite Systems (GNSS). The initial tilt angles (roll and pitch) are estimated by an extended Kalman filter (EKF) that uses two horizontal accelerometers as measurements. The initial position and heading estimation is performed using a subimage matching algorithm based on normalized cross-correlation between projected 2-D LiDAR scans and an occupancy floor map of the environment. The ambiguities in position/heading initialization are resolved using RSS. The proposed position/heading estimation module is also utilized in navigation mode as a source of absolute position/heading updates to EKF for enhanced observability. The state predictor is an enhanced 3-D inertial navigation system that utilizes low-cost microelectromechanical system (MEMS)-based reduced inertial sensor set aided by vehicle odometry. In navigation mode, LiDAR scans are used to estimate the vehicle's relative motions using an inertial-aided iterative closest point algorithm. To fuse all available measurements, a multirate multimode EKF design is proposed to correct navigation states and estimate sensor biases. The developed system was tested under a real indoor office environment covered by an IEEE 802.11 WLAN on a mobile robot platform equipped with MEMS inertial sensors, a WLAN interface, a 2-D LiDAR scanner, and a quadrature encoder. Results demonstrated the capabilities of the self-alignment and initialization module and showed average submeter-level positioning accuracy.

Asynchronous vehicle pose correction using visual detection of ground features

The inherent noise associated with odometry manifests itself as errors in localization for autonomous vehicles. Visual odometry has been previously used in order to supplement classical vehicle odometry. However, visual odometry is limited in its ability to reduce errors in localization for large travel distances that entail the cumulative summing of individual frame-to-frame image errors. In this paper, a novel machine vision approach for tiled surfaces is proposed to address this problem. Tile edges in a laboratory environment are used to define a travel trajectory for the Quansar Qbot (autonomous vehicle) built on the iRobot iRoomba platform with a forward facing camera. Tile intersections are used to enable asynchronous error recovery for vehicle position and orientation. The proposed approach employs real-time image classification and is feasible for error mitigation for large travel distances.The average position error for an 8m travel distance using classical odometry was measured to be 0.28m. However, implementation of the proposed approach resulted in an error of 0.028m. The proposed approach therefore significantly reduces pose estimation error and could be used to supplement existing modalities such as GPS and Laser-based range sensors.

Active optical sensors for tree stem detection and classification in nurseries.

Active optical sensing (LIDAR and light curtain transmission) devices mounted on a mobile platform can correctly detect, localize, and classify trees. To conduct an evaluation and comparison of the different sensors, an optical encoder wheel was used for vehicle odometry and provided a measurement of the linear displacement of the prototype vehicle along a row of tree seedlings as a reference for each recorded sensor measurement. The field trials were conducted in a juvenile tree nursery with one-year-old grafted almond trees at Sierra Gold Nurseries, Yuba City, CA, United States. Through these tests and subsequent data processing, each sensor was individually evaluated to characterize their reliability, as well as their advantages and disadvantages for the proposed task. Test results indicated that 95.7% and 99.48% of the trees were successfully detected with the LIDAR and light curtain sensors, respectively. LIDAR correctly classified, between alive or dead tree states at a 93.75% success rate compared to 94.16% for the light curtain sensor. These results can help system designers select the most reliable sensor for the accurate detection and localization of each tree in a nursery, which might allow labor-intensive tasks, such as weeding, to be automated without damaging crops.

Probabilistic Egomotion for Stereo Visual Odometry

We present a novel approach of Stereo Visual Odometry for vehicles equipped with calibrated stereo cameras. We combine a dense probabilistic 5D egomotion estimation method with a sparse keypoint based stereo approach to provide high quality estimates of vehicle’s angular and linear velocities. To validate our approach, we perform two sets of experiments with a well known benchmarking dataset. First, we assess the quality of the raw velocity estimates in comparison to classical pose estimation algorithms. Second, we added to our method’s instantaneous velocity estimates a Kalman Filter and compare its performance with a well known open source stereo Visual Odometry library. The presented results compare favorably with state-of-the-art approaches, mainly in the estimation of the angular velocities, where significant improvements are achieved.

Visual navigation system for small unmanned aerial vehicles

PurposeThis paper aims to address three major issues in the development of a vision‐based navigation system for small unmanned aerial vehicles (UAVs) which can be characterized as follows: technical constraints, robust image feature matching and an efficient and precise method for visual navigation.Design/methodology/approachThe authors present and evaluate methods for their solution such as wireless networked control, highly distinctive feature descriptors (HDF) and a visual odometry system.FindingsProposed feature descriptors achieve significant improvements in computation time by detaching the explicit scale invariance of the widely used scale invariant feature transform. The feasibility of wireless networked real‐time control for vision‐based navigation is evaluated in terms of latency and data throughput. The visual odometry system uses a single camera to reconstruct the camera path and the structure of the environment, and achieved and error of 1.65 percent w.r.t total path length on a circular trajectory of 9.43 m.Originality/valueThe originality/value lies in the contribution of the presented work to the solution of visual odometry for small unmanned aerial vehicles.

A Passive UHF-RFID System for the Localization of an Indoor Autonomous Vehicle

A global localization system combining odometry data with radio frequency identification (RFID) readings is proposed. RFID tags are placed at the ceiling of the environment and can be detected by a mobile robot unit traveling below them. The detection of the tags is the only information used in the proposed approach (no distance or bearing to the tag is considered available), but differently from similar localization setups reported in the literature, only a small number (about one each square meter or less) of tags are used. This is possible using a suitable tag's antenna in ultrahigh frequency band, expressly designed to obtain regular and stable RFID detection regions, which allows us to consider an efficient Kalman filtering approach to fuse RFID readings with the vehicle odometry data. A satisfactory performance is achieved, with an average position error of about 0.1 m. The hardware/software localization setup described in this paper is cheap and easy to use and may provide a satisfactory approach in several industrial and domestic scenarios.

Objective speckle velocimetry for autonomous vehicle odometry

Speckle velocimetry is investigated as a means of determining odometry data with potential for application on autonomous robotic vehicles. The technique described here relies on the integration of translation measurements made by normalized cross-correlation of speckle patterns to determine the change in position over time. The use of objective (non-imaged) speckle offers a number of advantages over subjective (imaged) speckle, such as a reduction in the number of optical components, reduced modulation of speckles at the edges of the image, and improved light efficiency. The influence of the source/detector configuration on the speckle translation to vehicle translation scaling factor for objective speckle is investigated using a computer model and verified experimentally. Experimental measurements are presented at velocities up to 80 mm s(-1) which show accuracy better than 0.4%.

Background Foreground Segmentation for SLAM

To perform simultaneous localization and mapping (SLAM) in dynamic environments, static background objects must first be determined. This condition can be achieved using a priori information in the form of a map of background objects. Such an approach exhibits a causality dilemma, because such a priori information is the ultimate goal of SLAM. In this paper, we propose a background foreground segmentation method that overcomes this issue. Localization is achieved using a robust iterative closest point implementation and vehicle odometry. Background objects are modeled as objects that are consistently located at a given spatial location. To improve robustness, classification is performed at the object level through the integration of a new segmentation method that is robust to partial object occlusion.

Correlation‐based visual odometry for ground vehicles

AbstractReliable motion estimation is a key component for autonomous vehicles. We present a visual odometry method for ground vehicles using template matching. The method uses a downward‐facing camera perpendicular to the ground and estimates the motion of the vehicle by analyzing the image shift from frame to frame. Specifically, an image region (template) is selected, and using correlation we find the corresponding image region in the next frame. We introduce the use of multitemplate correlation matching and suggest template quality measures for estimating the suitability of a template for the purpose of correlation. Several aspects of the template choice are also presented. Through an extensive analysis, we derive the expected theoretical error rate of our system and show its dependence on the template window size and image noise. We also show how a linear forward prediction filter can be used to limit the search area to significantly increase the computation performance. Using a single camera and assuming an Ackerman‐steering model, the method has been implemented successfully on a large industrial forklift and a 4×4 vehicle. Over 6 km of field trials from our industrial test site, an off‐road area and an urban environment are presented illustrating the applicability of the method as an independent sensor for large vehicle motion estimation at practical velocities. © 2011 Wiley Periodicals, Inc.

Vertically aligned accelerometer for wheeled vehicle odometry

A new application for accelerometers is presented and evaluated instead of using it to measure the linear acceleration of the vehicle, which is usually performed with traditional odometry. In the present case, a dual orthogonal axis accelerometer is vertically aligned in the center of a wheel. Then the rotational position of the wheel can be detected and RPM can be counted based on the accelerometer angular position. A microcontroller is used to measure the value of each accelerometer axis, and with a proper algorithm 12 wheel positions are detected combining the x-axis and y-axis values. The wheel position is transmitted by means of a wireless link to another microcontroller or computer located on board of the vehicle to estimate the traveled distance. This embedded wireless device is now named accelodometer. Field results on a pick-up tested at up to 40 km/h and with an autonomous robotic vehicle showed that this accelodometer presented advantages with respect to both usual accelerometer and incremental encoder, avoiding the accumulative error due to the sensor drift. The repeatability of the device was established within the 0.01% in 4 km path. A resolution of 1/12 of the wheel perimeter was determined.

Grid-based localization and local mapping with moving object detection and tracking

We present a real-time algorithm for simultaneous localization and local mapping (local SLAM) with detection and tracking of moving objects (DATMO) in dynamic outdoor environments from a moving vehicle equipped with a laser scanner, short-range radars and odometry. To correct the vehicle odometry we introduce a new fast implementation of incremental scan matching method that can work reliably in dynamic outdoor environments. After obtaining a good vehicle localization, the map surrounding of the vehicle is updated incrementally and moving objects are detected without a priori knowledge of the targets. Detected moving objects are finally tracked by a Multiple Hypothesis Tracker (MHT) coupled with an adaptive Interacting Multiple Model (IMM) filter. The experimental results on datasets collected from different scenarios such as: urban streets, country roads and highways demonstrate the efficiency of the proposed algorithm.