Free-Viewpoint Image Synthesis for Low-Bitrate VR Streaming

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing virtual reality technologies are limited by high data rates and restricted freedom of movement, failing to provide an immersive experience with real-life captured video due to the need for multiple video streams and high bitrates, and lack the flexibility to allow users to freely change viewing positions and directions.

Innovation Solution

A system that synthesizes images based on a stored database of real-life captured images, using a viewing position parameter vector to select and transform images, allowing users to dynamically change their position and direction with reduced bandwidth requirements by transmitting only one image stream, and adjusting for latency at the client side.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple cameras with different viewing positions are used to allow dynamic viewing position changes, then the freedom of viewing position selection is improved, but the data rate and system complexity increase significantly

Engineering Contradiction:
Improvefreedom of viewing position selectionVSAvoiddata rate
Core Design Contradiction:
Adaptability or versatilityVSQuantity of substance

Solution Approach 1:

The system segments the scene into multiple depth layers (foreground, midground, background) and processes each layer separately. This allows selective transmission of depth information only for relevant layers, reducing overall data rate while maintaining freedom of viewing position changes.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent adds a depth dimension to traditional 2D video by incorporating depth maps and performing depth-based rendering. This enables free-viewpoint video where users can change viewing positions without requiring multiple parallel video streams, as the depth information allows synthetic view generation from a single captured video sequence.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Adaptability or versatility

If multiple parallel video streams are distributed to provide viewing position changes, then the adaptability to user position is improved, but the device complexity and processing requirements increase

Engineering Contradiction:
Improveviewing position adaptabilityVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary depth extraction and layer segmentation during video capture or pre-processing. Depth maps are generated in advance for each depth layer, so that during playback only lightweight depth-based rendering operations are needed, avoiding complex real-time processing of multiple video streams.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Instead of transmitting multiple actual video streams from different cameras, the system creates synthetic copies of the scene by rendering 2D video frames with associated depth maps. These synthetic views are generated computationally rather than captured physically, reducing the number of required camera systems and data transmission requirements.

Inventive Principle:
Principle #26Copying

3Manufacturing precision

If depth maps and multiple view-points are encoded and transmitted, then the quality of three-dimensional virtual reality experience is improved, but the bitrate and data transmission requirements increase

Engineering Contradiction:
Improvethree-dimensional image qualityVSAvoidbitrate
Core Design Contradiction:
Manufacturing precisionVSQuantity of substance

Solution Approach 1:

The system applies different quality levels to different depth layers. Foreground objects with higher visual importance receive higher quality depth mapping and processing, while background objects use lower quality approximations. This selective quality allocation maintains overall 3D experience quality while reducing total bitrate requirements.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent transmits only the essential depth information needed for free-viewpoint rendering rather than complete three-dimensional models. By using depth maps from a limited number of captured views and interpolating intermediate depths, the system achieves adequate 3D quality with partial depth information, reducing bitrate compared to transmitting full multi-view stereo data.

Inventive Principle:
Principle #16Partial or excessive action

4Manufacturing precision

If free-viewpoint video with multiple view-points is implemented, then the virtual reality experience quality is improved, but the freedom of movement and viewing direction changes remains restricted

Engineering Contradiction:
Improvevirtual reality experience qualityVSAvoidfreedom of movement
Core Design Contradiction:
Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

By incorporating depth maps and performing depth-based compositing, the system adds a third dimension to traditional multi-view video. This enables continuous viewing position changes along the depth axis, not just discrete switching between pre-captured camera positions, thereby increasing freedom of movement within the virtual environment.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The system dynamically adjusts rendering parameters including viewing position, viewing direction, and depth layer composition based on user input. By changing these parameters in real-time and re-rendering the composite image from the desired viewpoint, the system provides continuous freedom of movement rather than restricted discrete position switching.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3520414B1Image processing
Publication Date: 2026.04.22 KONINKLIJKE PHILIPS NV
  • EP3520414B1 patent drawingFigure 1
  • EP3520414B1 patent drawingFigure 2
  • EP3520414B1 patent drawingFigure 3

AI summary

An apparatus comprises a store (201) storing images corresponding to different positions and viewing directions for a scene, and associated position parameter vectors for the images where the vector for an image comprises data indicative of a viewing position and a viewing direction. A receiver (205) receives a viewing position parameter vector from a remote client (101). A selector (207) selects a set of images in response to a comparison of the viewing position parameter vector and the associated position parameter vectors. An image synthesizer (209) generates an image from the set of images. A data generator (215) generates a reference position parameter vector for the synthesized image indicating a viewing position and direction for the synthesized image. An image encoder (211) encodes the synthesized image and an output generator (213) generates an output image signal comprising the encoded synthesized image and the reference position parameter vector. The output image signal is transmitted to the remote client.