Camera-Map Vehicle Localization Without Mono-Depth Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle localization methods for autonomous vehicles face challenges in accuracy, especially in scenarios with poor satellite connections and are prone to errors due to depth issues in mono-depth estimations, which affect the reliability of pose estimation and orientation.
Innovation Solution
A method that predicts a vehicle's pose based on sensor data and updates it using image-frame coordinate systems by transforming map road references into polylines, identifying corresponding features, projecting them, and determining error parameters to correct predictions, thereby improving accuracy without significant impact on size, power consumption, or cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If GNSS and IMU are used for positioning, then positioning coverage is improved, but measurement precision deteriorates due to large scale and bias errors
Solution Approach 1:
The patent introduces map road references as an intermediary element to bridge the vehicle and the positioning system. By transforming map road references into the image frame coordinate system and comparing them with detected road features, the system creates an intermediate reference framework that enables accurate positioning without relying solely on GNSS/IMU measurements, thus resolving the contradiction between coverage and precision
Solution Approach 2:
The patent replaces the mechanical/sensor-based positioning system (GNSS/IMU) with a vision-based geometric positioning approach. Instead of relying on physical sensors that suffer from drift and errors, the system uses image processing and geometric transformations of map data to determine vehicle pose, achieving higher measurement precision while maintaining broad coverage
2Device complexity
If mono-depth estimation is used to transform map data, then device complexity is reduced, but measurement precision deteriorates due to depth errors
Solution Approach 1:
The patent changes the dimension of transformation by working entirely in the image frame coordinate system rather than transforming to ego-vehicle coordinates. This dimensional approach eliminates the need for mono-depth estimation, as all transformations and comparisons occur in the 2D image plane where geometric relationships are preserved without requiring depth information, thus maintaining both low complexity and high precision
3Reliability
If HD-map and multiple sensors are combined, then reliability is improved, but device complexity increases
Solution Approach 1:
The patent merges the HD-map data with the camera imaging model by transforming map road references directly into the image frame coordinate system. This merging creates a unified representation where map features and image features exist in the same coordinate space, enabling direct comparison and validation without requiring complex integration of multiple sensor systems, thus achieving high reliability with controlled complexity
Data Source
AI summary
The present disclosure relates to a method for determining a vehicle pose, predicting a pose (xk, yk, θk) of vehicle on a map based on sensor data acquired by a vehicle localization system, transforming a set of map road references of a segment of a digital map from a global coordinate system to an image-frame coordinate system of a vehicle-mounted camera based on map data and predicted pose of the vehicle. The transformed set of map road references form a set of polylines in image-frame coordinate system. Identifying a set of corresponding image road reference features in an image acquired by vehicle mounted camera, where each identified road references feature defines a set of measurement coordinates (xi, yi) in image-frame. Projecting each of identified set of image road reference features onto formed set of polylines in order to obtain a set of projection points.