China
Africa
Explore chapters and articles related to this topic
Advanced Sensors and Vision Systems
Published in Marina Indri, Roberto Oboe, Mechatronics and Robotics, 2020
Visual odometry is the process of recovering the related position and orientation of a robot by analyzing the associated vision devices. Alternatively, visual odometry is defined as the estimation of the egomotion of the robot using only cameras mounted on the robot [20, 21]. It aims to recover the parameters of the equivalent odometry data using sequential camera images to estimate the motion of travel by the robot. The visual odometry technique allows enhancement of the navigation precision of robots in a variety of indoor or outdoor environments and terrain. An example of visual odometry estimates a robot's motion using camera images to extract relative features between two sequential images. Vision-based motion estimation is a very important task in applications of autonomous navigation of robots. The general flowchart of motion estimation based on vision sensors is illustrated in Figure 11.1.
Surveillance Systems and Applications
Published in Maheshkumar H. Kolekar, Intelligent Video Surveillance Systems, 2018
Video content analytics (VCA) is the capability to automatically analyze video to detect abnormal activities. This technical capability is used in video surveillance at public places such as airports, railway stations, and banks. The algorithms can be implemented as software on general purpose computers, or as developed dedicated hardware. Many different functionalities can be implemented in VCA, such as video motion detection. More advanced functionalities such as video object tracking and egomotion estimation can be included. It is also possible to build other functionalities in VCA, such as person identification, face detection, behavior analysis, and abnormal activitydetection.
A review on camera ego-motion estimation methods based on optical flow for robotics
Published in Lin Liu, Automotive, Mechanical and Electrical Engineering, 2017
As robotics rapidly develops in the field of automotive engineering, ego-motion estimation becomes one of the primary tasks to achieve robot control. The ego-motion of a robot is defined as the robot’s motion relative to the scene. Various types of sensors are invented for the purpose of ego-motion examination, and camera is the most common one among them.
Dense 3D surface reconstruction of large-scale streetscape from vehicle-borne imagery and LiDAR
Published in International Journal of Digital Earth, 2021
Xiaohu Lin, Bisheng Yang, Fuhong Wang, Jianping Li, Xiqi Wang
Visual-LiDAR SLAM: the integration of multi-modal measurements from camera and LiDAR is often addressed within a SLAM framework (Graeter, Wilczynski, and Lauer 2018). For example, Zhang, Kaess, and Singh (2014) associated depth information from LiDAR to visual features, resulting in a RGB-D system with augmented LiDAR depth. Shin, Park, and Kim (2018) used the depth from LiDAR in a direct method, where photometric errors were minimized in an iterative way to prevent local drift and maintain the consistency of the global. But direct method is sensitive to illumination changes. Recently, Zhang and Singh (2018) proposed a laser-visual-inertial odometry and mapping system which running three modules sequentially to produce high rate ego-motion estimation and low drift maps. However, these sophisticated pipelines need parallel processing to allow for real time performance. LIC-Fusion (Zuo et al. 2019) developed a tight coupled multi-modal sensor fusion algorithm for LiDAR-inertial-camera odometry within the multi-state constraint Kalman filter (MSCKF) framework. However, it needs efficient loop closure constraints obtained from LiDAR and camera to bound navigation errors. In order to overcome the failure of LOAM in the tunnel environment, VIL-SLAM (Shao et al. 2019) adopts the tight coupled stereo visual inertial LSLAM and the visual closed-loop detection with LiDAR enhancement. These methods can accurately and reliably estimate the pose of the vehicle, but it is sparse reconstruction due to the real-time demand, and few studies have been done on dense 3D reconstruction of large-scale streetscapes.
Stereo dense depth tracking based on optical flow using frames and events
Published in Advanced Robotics, 2021
Antea Hadviger, Ivan Marković, Ivan Petrović
In order to predict the dense disparity for any timestamp between the frames, it is required to determine the ego-motion of the camera. In all our experiments, we used the ground truth odometry provided in the datasets, but any ego-motion estimation method can be used. We assume constant motion of the camera in the whole interval until the next camera frame arrives, i.e. until the arrival of new ego-motion information. Thus we use the most recent ego-motion information as a prediction for the future camera motion. Disparity prediction can be done at an arbitrary frequency, depending on the available computational resources and required temporal resolution of dense disparity availability, It can depend on the event rate or the velocity itself, but it can be fixed as well. In any case, prediction is based on the dense disparity that has been continuously updated using events.
Improving monocular visual SLAM in dynamic environments: an optical-flow-based approach
Published in Advanced Robotics, 2019
Jiyu Cheng, Yuxiang Sun, Max Q.-H. Meng
An overview of our proposed method is shown in Figure 2. There are two modules in the proposed method. The first module called Ego-motion Estimation is to estimate the camera ego-motion between two consecutive frames j and j−1. Our proposed method works as follows: Firstly, two consecutive images are captured and denoted as RGB Curr and RGB Last, as shown in Figure 2. We employ the Five-Point Algorithm [42] to estimate the motion of the camera from the last image to the current image by computing the essential matrix. Finally we multiply the last image with the estimated transformation matrix to get a new image which we call the estimated image. In this case, points in the estimated image are converted to the current image. The second module, Dynamic Feature Points Detection, calculates optical flow value for each feature point extracted from the current image between the current image and the estimated image and detects dynamic feature points for the current image based on optical flow values. Static points are used for further camera pose estimations.