Real-Time Ray Tracing Lens Remapping for Automotive Surround View

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current automotive surround view camera systems face performance limitations due to the computational intensity of raster processing, particularly in generating 3D images with wide-angle cameras, as they require extensive GPU resources for rendering and lens remapping, which can be inefficient for complex scenes and dynamic views.

Innovation Solution

Implementing real-time ray tracing for lens remapping, which moves computationally expensive operations out of the per-image rendering loop and performs them offline during calibration, allowing for reduced complexity and increased performance by using ray-tracing enabled GPUs with hemispheric lens models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If raster processing with GPU is used to render 3D surround views, then the system can generate composite views of vehicle surroundings, but the GPU resources required scale exponentially with increased system complexity

Engineering Contradiction:
Improvesurround view featuresVSAvoidGPU resources
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent replaces traditional raster processing mechanics with ray tracing mechanics. Instead of using GPU-based rasterization pipelines that require exponential scaling of resources, the system employs ray tracing algorithms that cast rays from the viewpoint through lens models to determine pixel values, fundamentally changing the computational approach from mechanical raster scanning to optical path simulation

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the computational parameters by moving lens remapping operations from the per-image rendering loop to offline calibration. By pre-computing ray intersections with lens models and storing results in lookup tables, the system transforms real-time complex calculations into simpler interpolation operations during rendering, reducing runtime computational burden

Inventive Principle:
Principle #35Parameter changes

2Productivity

If raster processing is used for lens remapping in per-image rendering loop, then views can be rendered, but computational intensity reduces performance

Engineering Contradiction:
Improverendering performanceVSAvoidGPU cycles
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent applies preliminary action by performing lens remapping calculations during offline calibration before actual rendering. Ray intersections with lens models are pre-computed and stored, so that during runtime, only simple lookup and interpolation operations are needed, eliminating the need for intensive per-image lens remapping computations

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses copying by creating lookup tables that store pre-computed ray intersection data with lens models. Instead of recalculating complex lens remapping operations for each image, the system copies and interpolates values from these pre-computed lookup tables, dramatically reducing computational cycles required during rendering

Inventive Principle:
Principle #26Copying

Data Source

PatentUS11303807B2Using real time ray tracing for lens remapping
Publication Date: 2022.04.12 TEXAS INSTRUMENTS INC
  • US11303807B2 patent drawing
  • US11303807B2 patent drawing
  • US11303807B2 patent drawing

AI summary

A method for generating a surround view (SV) image for display in a view port of an SV processing system is provided that includes capturing, by at least one processor, corresponding images of video streams from each camera of a plurality of cameras and generating, by the at least one processor, the SV image using ray tracing for lens remapping to identify coordinates of pixels in the images corresponding to pixels in the SV image.