Stereo Vision Census Transform for Lower-Complexity Depth Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing depth sensing technologies face challenges in reducing computational complexity and memory usage when processing captured stereo images for generating depth data, particularly due to the intensive nature of census transforms applied to large amounts of image data.
Innovation Solution
Implementing a census transform that is performed only for pixels corresponding to known locations of a projected light pattern, such as a geometric pattern with known dot projections, reduces the amount of information generated and processed, thereby decreasing computational complexity and memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If census transform is performed for all pixels in stereo images, then depth data can be generated with complete coverage, but computational complexity and memory usage increase significantly
Solution Approach 1:
The patent extracts and processes only the subset of pixels that correspond to the projected light pattern, rather than processing all pixels in the image. This is achieved by identifying pixels at known locations where the light pattern is expected to appear, and performing census transform exclusively on these extracted pixels, thereby reducing computational complexity while maintaining depth measurement capability
Solution Approach 2:
The patent applies partial action by performing census transform on only a portion of the pixels (those corresponding to the light pattern) rather than all pixels. This partial processing approach reduces the computational burden and memory requirements while still generating sufficient depth data for the application
2Measurement precision
If census transform is performed for all pixels in stereo images, then complete depth information is obtained, but memory usage increases significantly
Solution Approach 1:
The patent extracts only the necessary pixel data corresponding to the projected light pattern for census transform processing. By identifying and processing only these specific pixels rather than all pixels in the stereo images, the quantity of data that needs to be stored in memory is significantly reduced, thereby lowering memory usage requirements
3Measurement precision
If census transform is performed on all pixels, then comprehensive depth data is generated, but processing time increases
Solution Approach 1:
The patent extracts and processes only the pixels corresponding to the projected light pattern, significantly reducing the number of pixels that require census transform processing. This extraction approach maintains comprehensive depth data generation for the regions of interest while substantially reducing the overall processing time
Solution Approach 2:
The patent applies partial action by performing census transform on only a subset of pixels rather than all pixels in the stereo images. This partial processing approach reduces the computational workload and corresponding processing time while still generating comprehensive depth data for the projected pattern regions
Data Source
AI summary
Systems and techniques are provided for wireless communication. For example, a process can include obtaining a first image of an environment, the first image including a pattern of light. The process can further include determining a pixel window for a pixel of the first image; determining a set of transform pixels based on the pattern of light, wherein the set of transform pixels are a subset of pixels from the pixel window; performing a census transform based on the pixel and the set of transform pixels to obtain census transform information for the first image; and generating depth data based on the census transform information for the first image and census transform information for a second image.