Depth-Map Feature Geometry Extraction for Lane-Level Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization technologies such as GPS, WiFi, and Bluetooth are imprecise for lane-level or road-level positioning due to multi-pathing, signal occlusion, and lack of precision in transmitting station locations, making them unreliable for determining the geographic location of devices.
Innovation Solution
Developing a fingerprint database using two-dimensional feature geometries extracted from depth maps, such as building facades and road signs, through linear, curvilinear, or machine learning algorithms, and encoding these geometries with geographic coordinates to enhance localization accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GPS, WiFi, or Bluetooth-based localization methods are used, then localization can be achieved, but positioning precision is insufficient to reach lane-level or road-level accuracy
Solution Approach 1:
The patent segments the environment into discrete visual features (geometric primitives) that can be independently detected and matched. By dividing the continuous visual scene into extractable features like lines, circles, and polygons, the system achieves precise localization without relying on traditional GPS/WiFi/Bluetooth methods, thereby resolving the contradiction between positioning precision and localization reliability.
Solution Approach 2:
The patent replaces the mechanical/electromagnetic signal-based localization systems (GPS, WiFi, Bluetooth) with a visual-based feature extraction and matching system. This substitution uses image processing and geometric feature recognition to achieve superior positioning precision at lane-level and road-level accuracy, eliminating the reliability issues inherent in signal-based methods.
2Measurement precision
If feature extraction and matching systems are implemented to achieve precise positioning, then positioning precision improves to lane-level accuracy, but computational requirements increase
Solution Approach 1:
The patent extracts only the essential geometric features from visual data, focusing on specific geometric primitives (lines, circles, polygons) rather than processing the entire image or all visual information. This selective extraction reduces computational complexity while maintaining lane-level positioning precision, as only critical features need to be identified and matched.
Solution Approach 2:
The patent transforms visual image data into geometric parameter representations (coordinates, dimensions, shapes of geometric primitives). By changing the parameter space from raw pixel data to simplified geometric features, the system achieves precise positioning with reduced computational requirements, as geometric parameter matching is more efficient than full image processing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems, apparatuses, and methods are provided for developing a fingerprint database and extracting feature geometries for determining the geographic location of an end-user device. A device collects, or a processor receives, a depth map of a location in a path network. A physical structure is identified within the depth map. The depth map is divided, at the physical structure, into a horizontal plane at an elevation from the road level. A two-dimensional feature geometry is extracted from the horizontal plane of the depth map using a linear regression algorithm, a curvilinear regression algorithm, or a machine learning algorithm.