Enriched Voxel Data Structure for Multi-Dimensional Scene Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data formats often focus on a single dimension, such as visual data, temporal information, or thermal imaging, failing to integrate multiple dimensions necessary for comprehensive understanding in real-world applications like autonomous vehicle navigation.
Innovation Solution
A data structure called daxle that encapsulates high-resolution images, accurate timestamps, thermal imaging layers, and 6 DOF tracking information within a single cohesive package, using enriched BOMs and neural networks to generate 2D pixels for a tensor.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing data formats focus on a single dimension (visual, temporal, or thermal), then the data structure remains simple, but the ability to comprehend complex real-world scenarios is insufficient
Solution Approach 1:
The patent combines multiple single-dimension data formats (visual images, temporal timestamps, thermal layers, 6DOF tracking) into a unified multi-dimensional data structure called daxle. This merging allows the system to comprehend complex real-world scenarios by integrating diverse data types while maintaining a coherent organizational framework through enriched bills of materials that link all data dimensions to specific spatial locations.
Solution Approach 2:
The daxle data structure serves multiple functions simultaneously: it stores visual imagery, temporal information, thermal data, and spatial tracking; it organizes data through enriched bills of materials; it enables comprehensive scene understanding; and it supports various downstream applications including autonomous navigation and environmental interpretation. This multi-functionality resolves the contradiction by making a single data structure adaptable to diverse computational needs.
2Loss of information
If multiple data dimensions are integrated into a unified structure, then comprehensive scene understanding is achieved, but the data processing complexity increases
Solution Approach 1:
The patent segments the complex multi-dimensional data into organized units called enriched bills of materials, which are further divided into data cubes representing specific spatial locations. Each data cube contains structured information about visual, thermal, temporal, and spatial characteristics at that location. This segmentation reduces processing complexity by breaking down the overwhelming multi-dimensional data into manageable, location-specific units that can be processed independently.
Solution Approach 2:
The enriched bill of materials acts as an intermediary structure that organizes and links multiple data dimensions (images, timestamps, thermal data, 6DOF tracking) to specific spatial locations. This intermediary layer simplifies the integration process by providing a standardized framework that connects diverse data types, reducing the complexity of directly managing multi-dimensional data relationships.
3Reliability
If diverse data streams are synthesized for holistic environmental representation, then interpretation accuracy in challenging conditions improves, but the computational resources required increase
Solution Approach 1:
The patent applies local quality by organizing data around specific spatial locations through enriched bills of materials and data cubes. Instead of processing all multi-dimensional data uniformly, the system focuses computational resources on specific locations of interest, creating high-resolution data representations only where needed. This approach improves interpretation accuracy in challenging conditions while reducing overall computational energy consumption by avoiding unnecessary processing of entire datasets.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Generating data enriched voxels is provided. The method comprises receiving image data of a three-dimensional (3D) object. A number of key vertices are detected within the 3D object, and a bill of materials (BOM) is created for each key vertex. The BOM for each key vertex is then enriched with production data and sensor data, wherein the enriched BOM for each key vertex describes environmental conditions within a defined area around the 3D object. The enriched BOM for each key vertex are then fed into a respective neural network that generates a two-dimensional (2D) pixel containing all data from the enriched BOM, wherein the 2D pixel forms part of a tensor of 2D pixels.