Band Matrix Compression for Faster Conjugate Gradient Depth Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for sparse matrix compression in depth fusion technology are inefficient, leading to high power consumption and inaccurate depth information due to large and sparse adjacency matrices, which hinder the use of depth sensors in mobile devices and complicate computations with the conjugate gradient method.
Innovation Solution
A conjugate gradient acceleration apparatus using band matrix compression, which converts adjacency matrices into band matrices, compresses data using run-length encoding, and performs computations efficiently through a band matrix calculation unit that reuses data without memory access, allowing for intermittent activation of depth sensors and improved power management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional sparse matrix compression is used, then memory usage is reduced, but compression efficiency remains low
Solution Approach 1:
The patent segments the sparse adjacency matrix into multiple band matrices based on depth information from the depth sensor. Each band matrix corresponds to a specific depth range, allowing independent compression and processing. This segmentation enables more efficient compression by exploiting the banded structure, achieving 32.6% higher compression efficiency while reducing memory usage compared to conventional sparse matrix compression methods.
2Productivity
If conventional sparse matrix computation is used, then computation can be performed, but data access to memory increases and computation speed decreases
Solution Approach 1:
The patent performs preliminary compression of the adjacency matrix into band matrices before the conjugate gradient computation. By pre-processing the matrix into a compressed banded format, the system reduces the amount of data that needs to be accessed during computation. This preliminary action minimizes memory access operations during the actual computation, thereby improving computation speed and reducing power consumption associated with memory access.
3Reliability
If depth sensor is continuously activated, then depth information is continuously available, but power consumption increases
Solution Approach 1:
The patent activates the depth sensor periodically rather than continuously, synchronizing sensor activation with the computational needs of the conjugate gradient method. The depth sensor is activated at specific intervals to provide depth information for matrix construction, then deactivated to conserve power. This periodic activation maintains sufficient depth information availability for accurate depth fusion while significantly reducing power consumption compared to continuous operation.
Data Source
AI summary
Disclosed is a conjugate gradient acceleration apparatus using band matrix compression in depth fusion technology including a band matrix conversion unit configured to convert an adjacency matrix for correcting depth data acquired from data of an image sensor through deep learning based on depth information acquired from a depth sensor into a band matrix using rows of the adjacency matrix as addresses of query points and columns of the adjacency matrix as the nearest neighbors at the query points, a band matrix compression unit configured to mark an index on each band in order to compress the band matrix and to compress data, a memory unit configured to store tile data of the band matrix, and a band matrix calculation unit configured to perform computation of the band matrix and a transposed band matrix or computation of a symmetric band matrix with respect to the band matrix and a vector.