Shared from my #HoloLens. Object detection, recognition, and measurement with HoloLens spatial mapping points (vertices). The CurvSurf's FindSurface SDK will be prepared for public distribution in 1Q 2017. Target OS: - Windows - Linux - Not yet Android. Augmented reality, least squares, orthogonal distance fitting, object tracking, object detection, object recognition, point cloud, plane, box, cuboid, sphere, cylinder, cone, torus, Microsoft Kinect, Google Project Tango, Intel RealSense, Microsoft HoloLens, LNCS 3151 Feb. 5, 2015 https://plus.google.com/+CurvSurf Real-Time Geometric Feature Extraction with 3-D Measuring Device Currently, 3-D measuring tablets are (will be) available on the market, e.g. Google Project Tango, Mantis Vision Aquila, DotProduct Phi3D, Intel RealSense R200, etc. CurvSurf will be developing software for real-time extraction of geometric primitives from point cloud with 3-D measuring device. 1. Automatic extraction of geometric primitives During or after a while of scanning, a point among the scanned point cloud is picked manually or automatically. Then, our software identifies the type of surface (plane, sphere, cylinder, cone, or torus) for a small patch of point cloud around the pick-point. The point cloud will be segmented and the surface parameters estimated by alternating region growing and model updating. Picking a point is repeated as long as no break condition is invoked. 2. Provision of pick-point The pick-point can be provided in various ways: a. The operator picks a point among the scanned point cloud by means of touching the device’s touch-screen. The displayed point next to the touch-point on screen is picked. b. The operator directs the device 1~2 seconds long onto an object point amid the scanned object surface region, then the displayed point next to the center of screen is picked. c. The operator gives a voice-command to the device, e.g. ‘plane!’ or ‘sphere!’ or ‘cylinder!’ or ‘cone!’ or ‘torus!’ or ‘any!’, then the displayed point next to the center of screen is picked. d. The operator gazes an object point and then an eye-tracker identifies the gazed point among the scanned object surface points. The gazed point is to be picked. e. The operator points an object point by using a pointing device or his finger and then a gesture-tracker identifies the pointed point among the scanned object surface points. The pointed point is to be picked. f. The color/intensity image of object surfaces is processed and segmented. The segmented image regions can be considered to represent individual object surfaces. The displayed points next to each center of the segmented image regions are picked. 3. Provision of surface-type The surface-type can be provided in various ways: a. The operator touches an icon on touch-screen, e.g. ‘plane’ or ‘sphere’ or ‘cylinder’ or ‘cone’ or ‘torus’ or ‘any’. b. The operator gazes an icon on screen or anywhere, e.g. ‘plane’ or ‘sphere’ or ‘cylinder’ or ‘cone’ or ‘torus’ or ‘any’, and then an eye-tracker identifies the gazed icon. c. The operator points an icon on screen or anywhere, e.g. ‘plane’ or ‘sphere’ or ‘cylinder’ or ‘cone’ or ‘torus’ or ‘any’, by using a pointing device or his finger and then a gesture-tracker identifies the pointed icon. d. The operator gives a voice-command to the device, e.g. ‘plane!’ or ‘sphere!’ or ‘cylinder!’ or ‘cone!’ or ‘torus!’ or ‘any!’. e. Through a curvature analysis for a small patch of point cloud around the current pick-point, four types of the object surface patch can be identified, i.e. planar, elliptic, parabolic, or hyperbolic patch: i. Planar patch may be a portion of plane. ii. Elliptic patch may be a portion of torus or sphere. If sphere is not explicitly given, torus is assumed. Sphere is a special instance of torus as spindle torus with a zero mean radius. iii. Parabolic patch may be a portion of cone or cylinder. If cylinder is not explicitly given, cone is assumed. Cylinder is a special instance of cone with a zero vertex angle. iv. Hyperbolic patch may be a portion of torus. 4. Region growing and model updating a. By analyzing the principal curvatures and principal axes for a small patch of point cloud around the pick-point, the initial parameter values of radii, vertex angle, location, and rotation of the given surface type as plane, sphere, cylinder, cone, or torus can be determined. b. By alternating region growing and model updating, the final parameter values of radii, vertex angle, location, and rotation of the given surface type as plane, sphere, cylinder, cone, or torus can be determined. c. If the cone obtained has a small vertex angle, a cylinder is assumed. d. If the torus obtained has a small mean radius, a sphere is assumed.
HoloLens: Feature Extraction by Voice Commands - YouTube |
| Likes | Dislikes |
| 3,197 views views | 377 followers |
| Science & Technology Creative Commons Attribution licence (reuse allowed) View attributions | Upload TimePublished on 8 Oct 2016 |
3d printer,3ds,3d ultrasound,3ds emulator,3d printing,3ds games,3ds xl,3d tattoos,3d pen,3d warehouse,3d pictures,3d tuning,3d nails,3d doritos,3ds max,3d shapes,3d printed house,3d scanner,3d printing service,3dmark
Không có nhận xét nào:
Đăng nhận xét