Homography Transformation Quantization for Faster Coordinate Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The generation of coordinate mapping matrices in homography transformation processes is slow due to the use of homography transformation matrices in float32 format, affecting image data processing efficiency.
Innovation Solution
Quantize the initial elements of the homography transformation matrix to a smaller data format, such as int8 or int16, to generate a quantized homography transformation matrix, which is used to create a coordinate mapping matrix, thereby reducing the byte count and improving generation speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If float32 format homography transformation matrix is used, then precision is maintained, but generation speed of coordinate mapping matrix is slow
Solution Approach 1:
The patent changes the data format parameter of the homography transformation matrix from float32 to quantized formats (int8, int16, or float16). This parameter change reduces the byte count from 4 bytes to 2 bytes or 1 byte per element, thereby improving the generation speed of the coordinate mapping matrix while maintaining sufficient precision for image data processing tasks.
2Manufacturing precision
If float32 format homography transformation matrix is used, then calculation accuracy is ensured, but memory space consumption is high
Solution Approach 1:
The patent applies parameter changes by quantizing the data format of the homography transformation matrix elements from float32 (4 bytes) to smaller formats such as int8 (1 byte), int16 (2 bytes), or float16 (2 bytes). This reduces the memory space consumption by 75% or 50% respectively, while the quantized values maintain sufficient calculation accuracy for practical image processing applications.
Data Source
Figure 1(a)~2
Figure 3
Figure 4~5
AI summary
Disclosed are a homography transformation-based image data processing method and apparatus, an electronic device, and a medium, including: obtaining an initial homography transformation matrix corresponding to initial image data in a first angle of view (S101), wherein each of initial elements in the initial homography transformation matrix occupies a first byte count of bytes; quantizing the initial elements in the initial homography transformation matrix to generate a quantized homography transformation matrix (S103), where each quantized element in the quantized homography transformation matrix occupies a second byte count of bytes, and the second byte count is less than the first byte count; generating a coordinate mapping matrix based on the quantized homography transformation matrix (S105); and further obtaining target image data in a second angle of view based on the initial image data and the coordinate mapping matrix (S107). In this case, generating the coordinate mapping matrix based on a quantized homography matrix that occupies a relatively small count of bytes after quantization saves space occupied by the quantized homography transformation matrix; and also improves a generation speed of the coordinate mapping matrix is improved, thereby improving overall efficiency of an image data processing process.