China
Africa
Explore chapters and articles related to this topic
AI for Autonomous Driving
Published in Josep Aulinas, Hanky Sjafrie, AI for Cars, 2021
Mono cameras tend to fall short in producing 3D information. Still, stereo camera systems can produce 3D information, although 3D reconstruction accuracy is constrained by image resolution, baseline length between the stereo-pair (i.e. the distance between the two cameras which produce two slightly differing images of the same thing), and the distance of the 3D object. By contrast, accurate 3D information is the main contribution of lidar. Lidar’s basic principle is to measure the time it takes for a light beam to travel from its source, bounce back from a surface and return to the source. Modern lidar “broadcasting” sensors can transmit hundreds of thousands of such light beams per second into predefined directions; their reflections produce a cloud of 3D points that indicate spatial position. Since it is assumed that each 3D point belongs to a certain object in the scene, after identification each point will also have to be “clustered”: assigned to the correct object.
Sensor Fusion in Multiscale Inspection Systems
Published in Wolfgang Osten, Optical Inspection of Microsystems, 2019
Flexible and fast inspection requires at first a way out of the known area of conflict of optical measurement systems, consisting of the measurement time, the field size, and the achieved resolution. Second, it is also desired to highly increase the application range concerning supported geometries, sizes of the objects, materials, and inspectable features. Solutions for these requirements are the integration of multiple sensors into one measurement system. The advantage can then be the acquisition of redundant or complementary information, a saving of time, or an improved quality of the acquired data [3]. In the literature, there are different schemes to describe such sensor fusion strategies. In addition to a chronological sequence or spatial arrangement of sensors, there is also a more general classification based on the overall fusion of information. This has been mainly proposed by Durrant-Whyte [2], Luo et al. [3], or Ruser et al. [4]. The corresponding sketch is given in Figure 17.1. Based on their findings, there are three different types of sensor and information fusion strategies: Competitive integration: Fusion of data from multiple uniform or homogeneous sensors, with the major objective to reduce the overall measurement uncertainty. By taking the mean value of redundant measurement data, it is possible to reduce parts of the uncorrelated and statistically distributed measurement uncertainty.Complementary integration: This is the fusion of different or uniform sensor data in order to fill information gaps. For instance, two sensors acquire different areas of an object to increase the overall field size. Another example is the acquisition of multimodal information by using different sensors such as the topography and the texture or color of an object.Cooperative integration: In the context of this fusion approach, information of various sensors has to be merged to get the desired result. This can be a stereo camera, where the three-dimensional (3D) information is reconstructed by two images, acquired by two separated image sensors. However, it is also possible to hierarchically combine different measurement devices. Here, information of a coarse sensor is used to guide a subsequent device to acquire high-precision measurement data [2]. Examples for this are multiscale sensor systems.
A calibration-free workflow for image-based mixed reality navigation of total shoulder arthroplasty
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2022
Wenhao Gu, Kinjal Shah, Jonathan Knopf, Chad Josewski, Mathias Unberath
In an effort to design a solution that minimises disruption to the surgical workflow, we propose a registration and navigation approach that is able to register anatomy in a texture-less surgical environment and guide drilling that overcomes human perception deficiency in depth and does not require intra-operative calibrations. We further evaluate the end-to-end pin placement accuracy on a phantom with both stereo image-based registration and marker-based registration methods and compare them with a patient-specific guide. Note that the marker attached to the stereo camera is to simplify the experiment and can be removed by attaching it rigidly to the HMD and calibrated with stereo camera calibration algorithm with HoloLens front facing camera. We first introduce the experimental setup in Sec. 2.1, followed by the reconstruction and registration process in Sec. 2.2. Finally, how the drill is guided is described in Sec. 2.3.
Extracting depth information from stereo images using a fast correlation matching algorithm
Published in International Journal of Computers and Applications, 2020
Mozammel Chowdhury, Junbin Gao, Rafiqul Islam
For a world 3D point P(xp,yp,zp), we can derive the equations for stereo dense depth estimation from the camera geometry model as follows: Our aim is to recover the 3D point P from its projections PL and PR. Therefore, we can get: Since the depth, zp indicates a distance value (i.e. in mm or cm), we need to modify Equation (3) for its uniformity because, the parameters (b, f, d) in the equation possess different units. Otherwise, it would provide erroneous result during depth or distance measurement between the stereo cameras and the object. Accordingly, we reform Equation (3) through converting the unit of the disparity value (d) by dividing it with the pixel size (normally in mm/pixel) of the camera. Thus the accurate depth information is given by, where s is the size of a pixel of the stereo camera. Therefore, once we estimate the disparity of a reference pixel, we can easily extract the depth information of that pixel.
Stereo camera visual SLAM with hierarchical masking and motion-state classification at outdoor construction sites containing large dynamic objects
Published in Advanced Robotics, 2021
Runqiu Bao, Ren Komatsu, Renato Miyagusuku, Masaki Chino, Atsushi Yamashita, Hajime Asama
In all experiments, a stereo camera was mounted on the cabin top of a roller facing the side. The baseline of the stereo camera was about 1 m. The roller moved along a typical trajectory (Figure 7(a)) with maximum speed of 11 km/h. The ground truth trajectories were recorded using RTK-GNSS. We synchronized ground truth and estimated camera poses by minimizing Absolute Trajectory RMSE [2,19,23] and choosing appropriate time offsets between GNSS's and the camera's timer. Then the estimated camera trajectories will be aligned with ground truth trajectories by Umeyama algorithm [24]. We evaluate the accuracy of camera pose outputs of the vSLAM system with reference to the associated ground truth by Absolute Trajectory RMSE (AT-RMSE).