Abstract: Described herein are methods and systems for remote visualization of real-time three-dimensional (3D) facial animation with synchronized voice. A sensor captures frames of a face of a person, each frame comprising color images of the face, depth maps of the face, voice data associated with the person, and a timestamp. The sensor generates a 3D face model of the person using the depth maps. A computing device receives the frames of the face and the 3D face model. The computing device preprocesses the 3D face model. For each frame, the computing device: detects facial landmarks using the color images; matches the 3D face model to the depth maps using non-rigid registration; updates a texture on a front part of the 3D face model using the color images; synchronizes the 3D face model with a segment of the voice data using the timestamp; and transmits the synchronized 3D face model and voice data to a remote device.

Abstract: Described herein are methods and systems for generating multiple maps during object scanning for 3D object reconstruction. A sensor device captures RGB images and depth maps of objects in a scene. A computing device receives the RGB images and the depth maps from the sensor device. The computing device creates a first map using at least a portion of the depth maps, a second map using at least a portion of the depth maps, and a third map using at least a portion of the depth maps. The computing device finds key point matches among the first map, the second map, and the third map. The computing device performs bundle adjustment on the first map, the second map, and the third map using the matched key points to generate a final map. The computing device generates a 3D mesh of the object using the final map.

Abstract: Described herein are methods and systems for three-dimensional (3D) object capture and object reconstruction using edge cloud computing resources. A sensor coupled to a mobile device captures (i) depth maps of a physical object, the depth maps including pose information, and (ii) color images of the object. An edge cloud device coupled to the mobile device via a 5G connection receives the depth maps and the color images. The edge cloud device generates a new 3D model of the object based on the depth maps and color images, when a 3D model of the object has not been generated. The edge cloud device updates an existing 3D model of the object based on the depth maps and color images, when a 3D model of the object has previously been generated. The edge cloud device transmits the new 3D model or the updated 3D model to the mobile device.

Abstract: Described herein are methods and systems for keyframe-based object scanning and tracking. A sensor device captures images of objects in a scene. For each image, a computing device labels each of at least a plurality of pixels in the image, tracks at least one region of the labeled image to determine an estimate of a current pose of at least one object, validates the estimate of the current pose of the at least one object, selects the labeled image as a keyframe based upon validation of the estimate of the current pose, and updates a volumetric model comprising the at least one object using the keyframe. The computing device generates a final 3D model of the at least one object based upon the updated volumetric model.

Abstract: Described herein are methods and systems for remote visualization of three-dimensional (3D) animation. A sensor of a mobile device captures scans of non-rigid objects in a scene, each scan comprising a depth map and a color image. A server receives a first set of scans from the mobile device and reconstructs an initial model of the non-rigid objects using the first set of scans. The server receives a second set of scans. For each scan in the second set of one or more scans, the server determines an initial alignment between the depth map and the initial model. The server converts the depth map into a coordinate system of the initial model, and determines a displacement between the depth map and the initial model. The server deforms the initial model to the depth map using the displacement, and applies a texture to at least a portion of the deformed model.

Abstract: Described are methods, systems, and apparatuses for 3D vision processing using an IP block. A vision processing module comprises an integrated circuit that performs one or more 3D vision processing algorithms and a plurality of controllers that couple the integrated circuit to each of: a sensor device, a processor, a memory module, and a network interface. The vision processing module receives image data from the sensor device, the image data corresponding to one or more images captured by the sensor device. The vision processing module executes one or more of the 3D vision processing algorithms using at least a portion of the image data as input. The vision processing module transmits an output from execution of one or more of the 3D vision processing algorithms to at least one of: the processor, the memory module, or the network interface.

Abstract: Described are methods and systems for real-time remote avatar creation and animation control. A sensor device captures images of non-rigid objects in a scene. A server coupled to the sensor device generates an initial 3D model for each of the non-rigid objects in the scene using the images. The server detects landmark points on the non-rigid objects using the initial 3D model. The server generates a control point animation map for the 3D model using the detected points. The server applies the animation map to the 3D model to generate a mapped 3D model. A viewer coupled to the server receives (i) the mapped 3D model and (ii) tracking information associated with the objects, including a model pose and deformation of the landmark points. The viewer modifies the mapped 3D model using the tracking information, and renders a video stream on the viewer using the modified 3D model.

Abstract: Described herein are methods and systems for dynamic real-time texture alignment for three-dimensional (3D) models. A computing device receives input images of objects in a scene, and generates a 3D model for at least one of the objects, comprising a plurality of mesh triangles. The computing device projects each mesh triangle of the 3D model to one of the input images. The computing device measures a texture discontinuity between adjacent mesh triangles of the projected image by comparing color differences in a shared edge of the adjacent mesh triangles. The computing device translates a texture associated with the adjacent mesh triangles in different directions to create texture candidates. The computing device applies the texture candidates to the corresponding mesh triangles until a seamless texture join is formed on the shared edge. The computing device generates a textured 3D model using the 3D model, the projected image, and the texture candidates.

Abstract: Described are methods and systems for generating a video stream of a scene including one or more objects. A sensor captures images of objects in a scene. A server coupled to the sensor, for each image, generates an initial 3D model for the objects and an initial 3D model of the scene. The server, for each image, captures pose information of the sensor as the sensor moves in relation to the scene or as the objects move in relation to the sensor. A viewing device receives the models and the pose information from the server. The viewing device captures pose information of the viewing device as the viewing device moves in relation to the scene. The viewing device renders a video stream on a display element using the received 3D models and at least one of the pose information of the sensor or the pose information of the viewing device.

Abstract: Described herein are methods and systems for texturing a three-dimensional (3D) model using photogrammetry. A sensor captures scans of a physical object, including related pose information, and color images corresponding to each scan. A computing device generates a 3D mesh of the physical object. The computing device preprocesses the color images to remove blurry images and detect textured regions of the object in each non-blurry image. The computing device optimizes the pose information for each color image by generating associations between the color images and vertices in the 3D mesh and classifying the vertices as textured or non-textured. The computing device generates texture coordinates for the 3D mesh by segmenting the mesh, parameterizing the segments, and packing the parameterized segments into a texture atlas. The computing device paint the texture atlas using the color images that have optimized pose information to generate a model having texture coordinates for each vertex.

Abstract: Described herein are methods and systems for closed-form 3D model generation of non-rigid complex objects from scans with large holes. A computing device receives (i) a partial scan of a non-rigid complex object captured by a sensor coupled to the computing device; (ii) a partial 3D model corresponding to the object, and (iii) a whole 3D model corresponding to the object, wherein the partial 3D scan and the partial 3D model each includes one or more large holes. The device performs a rough match on the partial 3D model and changes the whole 3D model using the rough match to generate a deformed 3D model. The device refines the deformed 3D model using a deformation graph, reshapes the refined deformed 3D model to have greater detail, and adjusts the whole 3D model according to the reshaped 3D model to generate a closed-form 3D model that closes holes in the scan.

Abstract: Methods and systems are described for generating a three-dimensional (3D) model of a fully-formed object represented in a noisy or partial scene. An image processing module of a computing device receives images captured by a sensor. The module generates partial 3D mesh models of physical objects in the scene based upon analysis of the images, and determines a location of at least one target object in the scene by comparing the images to one or more 3D reference models and extracting a 3D point cloud of the target object. The module matches the 3D point cloud of the target object to a selected 3D reference model based upon a similarity parameter, and detects one or more features of the target object. The module generates a fully formed 3D model of the target object using partial or noisy 3D points from the scene, extracts the detected features of the target object and features of the 3D reference models that correspond to the detected features, and calculates measurements of the detected features.

Abstract: Described herein are methods and systems for closed-form 3D model generation of non-rigid complex objects from incomplete and noisy scans. An image processing module receives a scan of a non-rigid complex object captured by a sensor, a 3D model corresponding to the object, and a camera trace, where the scan includes one or more holes. The module cleans the scan and the 3D model using the camera trace. The module deforms the cleaned 3D model to the cleaned scan and matches the deformed 3D model to the cleaned scan. The module determines one or more portions of the deformed 3D model that are unmatched and deforms the unmatched portions of the deformed 3D model to the scan using the matched portions of the deformed 3D model to generate a closed-form 3D model that closes the holes in the scan.

Abstract: Methods and apparatuses are described for processing 3D vision algorithms. A 3D vision processor device comprises one or more 3D vision processing cores. Each 3D vision processing core includes one or more memory blocks for storing location values associated with 3D point cloud images and an arithmetic logic unit coupled to the one or more memory modules. The arithmetic logic unit includes a plurality of memory registers for temporarily storing location values associated with a point in a 3D point cloud image and a processing unit coupled to the plurality of memory registers for performing arithmetic operations on the location values stored in the memory registers, the arithmetic operations used for 3D vision processing algorithms. The 3D vision processing core also includes a communication link for transferring data between the arithmetic logic unit and the memory modules.

Abstract: Described are computer-based methods and apparatuses, including computer program products, for compressing three dimensional data of a scene. Data is received comprising (i) three dimensional data of a scene, and (ii) depth data associated with the three dimensional data. A triangle mesh is generated based on the three dimensional data, the triangle mesh comprising a plurality of triangles, each triangle including three vertices and three edges connecting the three vertices. For each edge in the triangle mesh, a metric is calculated for the edge based on data from the depth data associated with the edge, a length of the edge, and a curvature of the edge. A set of edges is collapsed based on a metric associated with each edge in the set of edges to generate a compressed triangle mesh.

Abstract: A method of interpolating images from original images are disclosed. Interpolation of images is based on the captured images, normal vector of the image segment plane and distance between the camera and the original image. Image segment groups retrieved are selected in order to discard and select only the image segment groups that meet criteria. Further processing of applying a shape change, merging the interpolated images and smoothing of the merged images provide interpolated images for viewing at angle of a virtual camera.

Abstract: Described are computer-based methods and apparatuses, including computer program products, for compressing three dimensional data of a scene. Data is received comprising (i) three dimensional data of a scene, and (ii) depth data associated with the three dimensional data. A triangle mesh is generated based on the three dimensional data, the triangle mesh comprising a plurality of triangles, each triangle including three vertices and three edges connecting the three vertices. For each edge in the triangle mesh, a metric is calculated for the edge based on data from the depth data associated with the edge, a length of the edge, and a curvature of the edge. A set of edges is collapsed based on a metric associated with each edge in the set of edges to generate a compressed triangle mesh.