CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is a non-provisional of United States Provisional Patent

Application No. 62/347,077 filed June 7, 2016, the contents of which are hereby incorporated by reference.

BACKGROUND

[0002] Streaming 360-degree video content may provide immersive environments for virtual reality (VR) and augmented reality (AR) applications.

SUMMARY

[0003] In an aspect, an imaging system is provided. The imaging system includes a plurality of cameras configured to capture video image data based on respective fields of view of an environment. Each camera of the plurality of cameras is communicatively coupled to neighbor cameras of the plurality of cameras via a communication interface. Each camera of the plurality of cameras includes at least one processor and a memory. The at least one processor executes instructions stored in memory so as to carry out operations. The operations include capturing video image data of the respective field of view and determining an overlay region. The overlay region includes an overlapping portion of video image data captured by the respective camera and at least one of the neighbor cameras. The operations also include cropping and warping the captured video image data of the respective field of view based on the overlay region to form respective processed video image data.

[0004] In an aspect, a method is provided. The method includes receiving processed video image data associated with respective cameras of a plurality of cameras of an imaging system. Each camera of the plurality of cameras is configured to capture video images of respective fields of view of an environment. The processed video image data includes cropped and warped video image data based on an overlay region. The overlay region includes an overlapping portion of video image data captured by at least two neighbor cameras of the plurality of cameras. The method also includes providing streamed video to a client device, via a plurality of communication links. The streamed video is based on the processed video image data.

[0005] In an aspect, a system is provided. The system includes various means for carrying out the operations of the other respective aspects described herein. [0006] These as well as other embodiments, aspects, advantages, and alternatives will become apparent to those of ordinary skill in the art by reading the following detailed description, with reference where appropriate to the accompanying drawings. Further, it should be understood that this summary and other descriptions and figures provided herein are intended to illustrate embodiments by way of example only and, as such, that numerous variations are possible. For instance, structural elements and process steps can be rearranged, combined, distributed, eliminated, or otherwise changed, while remaining within the scope of the embodiments as claimed.

BRIEF DESCRIPTION OF THE FIGURES

[0007] Figure 1 illustrates a conventional virtual reality 360° video streaming pipeline.

[0008] Figure 2 illustrates a conventional virtual reality 360° video streaming pipeline.

[0009] Figure 3 illustrates a conventional virtual reality 360° video streaming pipeline.

[0010] Figure 4 illustrates a conventional virtual reality 360° video streaming pipeline.

[0011] Figure 5 illustrates a distributed virtual reality 360° video capture, processing, and streaming system, according to an example embodiment.

[0012] Figure 6 illustrates system, according to an example embodiment.

[0013] Figure 7 illustrates system with a detached camera, according to an example embodiment.

[0014] Figure 8 illustrates geometrical camera arrangement, according to an example embodiment.

[0015] Figure 9A and 9B illustrate inter-unit connections, according to an example embodiment.

[0016] Figure 10 illustrates warping and cropping in adjacent camera units, according to an example embodiment.

[0017] Figures 11A and 11B illustrate warping and cropping of image frames by estimating a depth map, according to an example embodiment.

[0018] Figure 12 illustrates warping and cropping of image frames by graph cut, according to an example embodiment.

[0019] Figure 13 illustrates a live stream from capturing to consumption, according to an example embodiment.

[0020] Figure 14 illustrates a live stream from capturing to consumption, according to an example embodiment.

DETAILED DESCRIPTION

[0021] Example methods, devices, and systems are described herein. It should be understood that the words "example" and "exemplary" are used herein to mean "serving as an example, instance, or illustration." Any embodiment or feature described herein as being an "example" or "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments or features. Other embodiments can be utilized, and other changes can be made, without departing from the scope of the subject matter presented herein.

[0022] Thus, the example embodiments described herein are not meant to be limiting.

Aspects of the present disclosure, as generally described herein, and illustrated in the figures, can be arranged, substituted, combined, separated, and designed in a wide variety of different configurations, all of which are contemplated herein.

[0023] Further, unless context suggests otherwise, the features illustrated in each of the figures may be used in combination with one another. Thus, the figures should be generally viewed as component aspects of one or more overall embodiments, with the understanding that not all illustrated features are necessary for each embodiment.

I. Overview

[0024] Virtual reality (VR) 360° video has a long pipeline from production to consumption. Figure 1 illustrates a conventional virtual reality 360° video streaming pipeline 100. (1) It starts with a VR 360° camera, which often consists of multiple cameras. These cameras are mounted to a rig and each camera is arranged to capture a different view angle. These cameras include overlapping fields of view (FOV) to cover the whole 360 x 180 degree spherical view by image stitching. (2) Data from every camera is read into a processing unit, which aligns and stitches multiple video streams into one 360° stream; (3) the 360° stream then may be edited; (4) the edited video is then encoded and uploaded to cloud; (5) the video is then transcoded in the cloud and be streaming-ready; (6) video contents are distributed to clients; and (7) client applications receive the contents and render/display them for end-users.

[0025] The pipeline may break into a few components in real implementation. For example, Figure 2 shows a conventional virtual reality 360° video streaming pipeline 200, which consists of A. a 360° camera, B. a powerful computer, C. a powerful cloud server, and D. a VR headset or smartphone. In this configuration. The powerful computer (C) does a lot of heavy work from stitching, editing, encoding, to uploading. Note that a computer receives a huge amount of data from the 360° camera, that stitching and encoding are expensive in computation, that the local storage has to be large enough to accommodate the large data, and that it will take lots of time and bandwidth to upload the processed video to a cloud in Internet. Because of these difficulties, this computer has to be made super powerful (Fast I/IO, Lots of memory, Fast CPU or GPU, and fast internet connection), which is expensive. More than that, while the 360° camera can easily scale up to include more cameras for higher quality, it is hard to scale up the powerful computer accordingly. As a result, this solution is usually not able to process the data fast enough for live streaming.

[0026] Figure 3 illustrates a conventional virtual reality 360° video streaming pipeline

300, in which encoding and uploading are done with dedicated device(s), and the powerful compute only deals with video editing and stitching. But again, as the 360° camera scales up, stitching may not be done fast enough for live streaming; and also one dedicate device may not be able to encode high resolution (e.g. 4K, 8K, 12K) video in real time.

[0027] Another a conventional virtual reality 360° video streaming pipeline 400 is illustrated in Figure 4. Stitching and encoding are done with a hardware solution in the 360° camera, and the powerful computer does further editing and encoding (if necessary) and uploads the final results. In this configuration, the work in the VR camera is often so heavy that it often has to sacrifice video quality and resolution.

[0028] We propose a distributed infrastructure for VR 360° video capturing and live streaming. Figure 5 illustrates a distributed virtual reality 360° video capture, processing, and streaming system 500, according to an example embodiment. Each camera of the system is integrated with processors (CPU, GPU), storage, and transmission units. According, each camera may be incorporated into an integrated module. This module may be configured to carry out image capturing, warping, stitching, editing, encoding, transcoding, to uploading, for a small FOV. This VR 360° video camera system consists of K integrated camera modules. These modules communicate with each other for best stitching, and may perform operations in parallel.

[0029] Since videos from each integrated camera module have been carefully warped and cropped, no more alignment is needed to stitch them together. One can choose to stitch all videos together in the cloud with relatively low computation cost, or choose to stitch them during rendering on the final display device (e.g., VR headset, smartphone, etc).

II. Example Systems

A. Integrated Camera Unit

[0030] In one embodiment, the integrated unit of camera, processor, storage, and transmission modules can be made into one piece of module, as shown in Figure 6. Namely, Figure 6 illustrates system 600, according to an example embodiment. [0031] In another embodiment, the camera may be physically separated from other components, as shown in Figure 7. Figure 7 illustrates a system 700 with a detached camera, according to an example embodiment. Such an arrangement may make it easier to attach cameras to the right position on the VR 360° camera rig.

[0032] Each camera unit may also include a microphone, so that all camera units together are able to record and stream sound in different directions.

B. Geometry of VR 360° Camera System

[0033] A VR 360° camera system may consist of various numbers of cameras. These cameras can be geometrically arranged in various ways to cover the desired FOV Figure 8 illustrates a geometrical camera arrangement 800, according to an example embodiment. Figure 8 shows an exemplar geometry arrangement of 14 camera units that cover a full sphere of view. Camera 1 on the top (Plane 1) looks upward and covers 100° x 100° FOV with a fisheye lens. Cameras 2-7 in Plane 2 look outward and cover a cylinder view of 50° x 360° FOV, each of them covering 50° x 65° FOV with 5° horizontal overlapping between consecutive (e.g., adjacent and/ or neighboring) cameras. The bottom half of the system (e.g., cameras 8-14) may be symmetric with respect to the top half.

[0034] Fewer cameras may be implemented if a FOV smaller than sphere is desired, or if each individual camera has a larger FOV. More cameras may be implemented if more overlap between cameras is desired (e.g., for easier stitching, more redundancy), or if each individual camera has a smaller FOV.

[0035] In an example embodiment, a pair of cameras may be arranged along each plane

(e.g., at each viewpoint) to provide a stereoscopic view for every view.

[0036] Also, although this disclosure provides examples involving 360° video, the same method and systems may be applied to provide videos with fields of view that are less than 360°.

C. Interconnection Between Camera Units

[0037] In the proposed VR 360° camera system, each camera unit processes video frames in a way that output frames from different camera units that may be directly stitched to form a spherical view, or may be stitched to form a spherical view with a small amount of further processing. In such a scenario, each camera needs to receive information from its neighboring camera units (e.g., neighbor cameras) to perform the image processing (e.g., warping and/or cropping).

[0038] In one embodiment, these camera units are directly connected to one another via a wired or wireless communication interface (e.g., BLUETOOTH, BLUETOOTH LOW ENERGY, WiFi, or another type of communication protocol), as shown in Figure 9A. Figure 9A illustrates inter-unit connections. In another embodiment, each camera unit may be connected to a center hub device, via a wired or wireless communication interface as shown in Figure 9B. This interconnection allows camera units to exchange information for respective image processing on each camera unit at runtime.

[0039] In one embodiment, a geometric position of every camera may be pre-calibrated, which may avoid communication between cameras during runtime. This simplifies the system at the cost of stitching quality. For example, high quality stitching is a result of both the geometric arrangement of cameras as well as the spatial position of imaged objects.

[0040] Batteries may or may not be included in the integrated units. For example, the imaging system may be externally powered. Additionally or alternatively, batteries may provide some or all power for the imaging system.

[0041] In an example embodiment, network devices may be incorporated in the communication link and hardware architecture between the VR camera system and the cloud server so as to speed up or help facilitate the uploading process.

III. Example Methods

A. Camera Synchronization

[0042] Each camera unit in the proposed VR 360° camera system captures a portion of a spherical view. To stitch images from each camera together into a 360° image frame (and a 360° video), image capture in each camera unit needs to be synchronized. Camera system clocks may be synchronized based on communication between the cameras. Additionally or alternatively, a synchronizing flash may be fired while the cameras are capturing video. In such a scenario, the cameras may be synchronized by finding the frames from the respective video data that capture the synchronizing flash. Additionally or alternatively, the cameras may be synchronized by analyzing the final video clips (e.g., by stopping the video capture of the cameras at the same time).

[0043] If all cameras are synchronized, then for any given time, t, in the target 360° video, one may locate one frame on each video clip that is closest to t, and stitch them together. Linear interpolation may be used here in the temporal dimension for better smoothness.

B. Warping and Cropping for Video Stitching

1. Background

[0044] Video stitching difficulty may stem from 1) lens distortion; and 2) field of view disparity between cameras. Lens distortion may be largely corrected by camera calibration which may be done before runtime or at runtime. Camera disparity is scene dependent, and may be addressed using camera overlaps when video is being captured.

[0045] Frames from one camera unit overlap with those from its neighbor camera units.

Figure 10 illustrates two frames from different camera units. Overlapped regions are marked using red rectangles. These overlapping regions are used to warp and crop frames. Once such warping and/or cropping has been performed on the individual frames, the frames can be stitched seamlessly as illustrated.

[0046] Information of overlapping regions may be transmitted in various ways. 1) One may transmit to its neighbor camera unit the max possibly overlay regions (as shown in red rectangles in Figure 10; 2) One may detect feature points (e.g., scale-invariant feature transform (SIFT) points) in these regions, and transmit these feature points to its neighbor camera unit; 3) One may transmit both overlap regions of pixels and detected feature points; 4) One may transmit these information for every frames, or a periodic or aperiodic interval of frames. For example, information about overlap regions may be transmitted to neighbor cameras in response to the captured scene changing by a threshold amount (e.g., based on an image histogram, or another image metric) and/or a feature near the known overlap region moving by more than a threshold distance.

2. Video Warping and Cropping for Each Camera Unit

[0047] Figures 11A and 11B illustrate warping and cropping of image frames by estimating a depth map, according to an example embodiment. Figures 11A and 11B show an example in a 2D view where five cameras uniformly distribute in the 360° view, each covers 90° FOV. In one method, each camera unit is responsible for producing a final image of 72° from the point of view of center point, so that combining all frames from five cameras yields one frame of 360° from a single point of view, O. The responsible FOV for each camera unit is pre-determined during calibration.

[0048] As shown in Figure 11B, with an image, Fl, from point of view A, and images

F2 and F3, from neighbor cameras from Point B and C, respectively, there are various ways to produce an image from a single point of view, O. An example process is shown in Figure 11B:

[0049] 1. Between Fl and F2, compute the disparities for the rightmost pixel of Fl, PI, and then infer the depth of this pixel.

[0050] 2. Between Fl and F3, compute the disparities for the leftmost pixel of Fl, P2, and then infer the depth of this pixel.

[0051] 3. Estimate depth of other pixels in Fl, use a linear interpolation of the depths of

PI and P2. [0052] 4. With depth of every pixel being estimated, we can remap Fl to the targeted

72° for the point of view O.

[0053] One need not compute depth explicitly to produce targeted images. Another exemplar solution is as shown in Figure 12. Figure 12 illustrates warping and cropping of image frames by graph cut, according to an example embodiment.

[0054] 1. For each frame from Point A (2), take the possibly overlapping regions. For example, Region W and U from Viewpoint C and A, and Region P and Q from Viewpoint A and B.

[0055] 2. For each pair of overlapping region, find the best cut that minimizes discontinuities between W and U, and P and Q, as shown in (4) and (5). A number of graph cut algorithms have been proposed in literature.

[0056] 3. As shown in (6), from Frame 1, crop off Region W and Region Q, and then horizontally warp the rest region into a rect of image.

[0057] 4. In reality, each frame may have four overlapping regions with its neighbor cameras. As a result, Frame 1 may be cropped in four directions, as shown in (7), W, V, Q, Y. The rest of the regions are then warped horizontally and vertically into the final rectangle image.

[0058] 5. Cropped and warped frames from all cameras will cover the full spherical view with a reduced amount of stitching artifacts.

[0059] In one embodiment, for temporal smoothness, extra smoothness constraints (e.g. , performing edge alignment, movement compensation, color/shape matching, tone mapping, exposure correction, etc.) may be posed when finding the best cut between overlapping regions.

[0060] This processing may be done for every frame or for a periodic or aperiodic interval of frames. The graph cut (e.g., cropping and warping) may be interpolated based on an estimated movement rate and/or other change in the respective images between image processing periods.

C. Data Storage

[0061] Video data from each camera unit may be saved immediately to local storage, or may also be cropped and warped first (as described in the previous session) before saving to a local storage.

[0062] Video data may also be encoded (e.g., H.264, VP8, VP9, etc) before saving to one file in a local storage. Video data may also be encoded into a series of small trunk files in a local storage for later streaming (e.g., HLS, DASH). D. Video Data Upload and Live Stream

[0063] Regardless of saving to a local storage or not, the processed video data may be uploaded to a cloud in real time. For an example as shown in Figure 13, as video data is encoded into small trunks, each trunk may be uploaded to a cloud immediately. Figure 13 illustrates a live stream 1300 from capturing to consumption, according to an example embodiment.

[0064] In this case, the cloud may be able to broadcast to a number of users via various streaming protocols (e.g., HLS, DASH, etc).

[0065] Note that the VR 360° camera system consists of a number of camera units. Each unit uploads one stream of data to the cloud as shown in Figure 13. A metadata may be created in the cloud stream server to describe the geometric information of the VR 360° camera system and other video metadata (e.g., framerate, resolution, video format, video/audio codec, etc).

1. Stream and Stitch on Client Application

[0066] An application on a client device first grabs the metadata from stream server, and then connects to all the required lists of video trunks. The application is designed to stream and synchronize the required video trunks as needed, stitch them together, render to screen, and so it provides a VR 360° video for end-users. One possible solution of application is described in US Provisional Patent Application No. 62/320,451, filed April 8, 2016.

2. Stitch on Cloud and then Stream to Client Application

[0067] One may also stitch video data from all camera units on the cloud. Since all data have already been aligned before uploading, computation for stitching is relatively low and may be done in real-time. After stitching, the stitched video data appears to be a regular video stream, which may be streamed to client devices via regular streaming protocols (e.g., HLS, DASH, etc).

3. Combine Cloud Stitching and Client Stitching

[0068] One may also stitch a low-resolution 360° video on the cloud, and stitch high- resolution 360° video in the client application. To save computation on the cloud, each camera unit in the VR 360° camera system may upload two series of trunks to the cloud, one high-res trunks and one low-res trunks, as shown in Figure 14. Figure 14 illustrates a live stream 1400 from capturing to consumption, according to an example embodiment.

[0069] The particular arrangements shown in the Figures should not be viewed as limiting. It should be understood that other embodiments may include more or less of each element shown in a given Figure. Further, some of the illustrated elements may be combined or omitted. Yet further, an illustrative embodiment may include elements that are not illustrated in the Figures. [0070] A step or block that represents a processing of information can correspond to circuitry that can be configured to perform the specific logical functions of a herein-described method or technique. Alternatively or additionally, a step or block that represents a processing of information can correspond to a module, a segment, or a portion of program code (including related data). The program code can include one or more instructions executable by a processor for implementing specific logical functions or actions in the method or technique. The program code and/or related data can be stored on any type of computer readable medium such as a storage device including a disk, hard drive, or other storage medium.

[0071] The computer readable medium can also include non-transitory computer readable media such as computer-readable media that store data for short periods of time like register memory, processor cache, and random access memory (RAM). The computer readable media can also include non-transitory computer readable media that store program code and/or data for longer periods of time. Thus, the computer readable media may include secondary or persistent long term storage, like read only memory (ROM), optical or magnetic disks, compact-disc read only memory (CD-ROM), for example. The computer readable media can also be any other volatile or non-volatile storage systems. A computer readable medium can be considered a computer readable storage medium, for example, or a tangible storage device.

[0072] While various examples and embodiments have been disclosed, other examples and embodiments will be apparent to those skilled in the art. The various disclosed examples and embodiments are for purposes of illustration and are not intended to be limiting, with the true scope being indicated by the following claims.