Panoramic Video Transcoding via GPU-CPU Task Splitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current panoramic video transcoding methods consume significant computing resources, making it difficult to achieve real-time processing and unable to meet the requirements for VR video live broadcasting due to high bit rate usage, even though they maintain high resolution at all visual angles, including those invisible to users.
Innovation Solution
The method involves using a CPU-GPU combination to decode, map, and encode panoramic video data, distributing these processes between the CPU and GPU to optimize resource utilization, with the GPU handling the mapping and encoding of multiple channels and the CPU handling slicing and packing, allowing for real-time transcoding by allocating tasks based on processing capacity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If panoramic video is transcoded maintaining the same resolution at all visual angles, then the definition and resolution are maintained, but the bit rate becomes excessively high due to processing invisible areas
Solution Approach 1:
The patent applies local quality by differentiating resolution requirements across different visual angles. The spherical panoramic video is divided into multiple visual angle regions, with high resolution allocated to the visible area (within user's field of view) and reduced resolution to invisible areas. This resolves the contradiction by making resolution quality local rather than uniform, maintaining definition where needed while reducing bit rate in unnecessary regions.
2Reliability
If traditional transcoding methods are used for panoramic video, then comprehensive processing is performed, but computing resources are excessively consumed and real-time processing cannot be achieved
Solution Approach 1:
The patent extracts and removes the unnecessary processing of invisible area data from the transcoding pipeline. By identifying and eliminating the redundant computation associated with visual angles outside the user's field of view, the system reduces computing resource consumption while maintaining the reliability of processing for visible content. This extraction enables real-time processing capability.
Solution Approach 2:
The patent segments the panoramic video processing into distinct visual angle channels, separating visible and invisible regions. This segmentation allows independent processing of different regions with appropriate computational resources, improving overall processing efficiency and enabling real-time performance while maintaining comprehensive processing reliability for relevant content.
3Measurement precision
If high resolution is maintained across all visual angles, then video definition is preserved, but resource waste occurs in invisible areas reducing transcoding efficiency
Solution Approach 1:
The patent implements local quality by applying different resolution levels to different spatial regions of the panoramic video. High definition is concentrated in the visible visual angles where users actually view content, while invisible areas use reduced resolution. This optimizes computing resource utilization by eliminating waste in invisible regions while preserving video definition where it matters.
Data Source
AI summary
Panoramic video transcoding methods, devices, and apparatuses are provided. The method involves a central processing unit (CPU) and a graphics processing unit (GPU) and includes decoding a panoramic video; mapping, at least partly using the GPU, the decoded video data to N visual angles, to obtain N channels of video data, in which N is more than 1; encoding the N channels of video data respectively to obtain N channels of video streams; and slicing and packing the N channels of video streams respectively.