Vehicle Localization Using Stationary Objects and Multi-Camera Orientation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle localization systems rely heavily on satellite navigation data, which can be inaccurate and fail to provide reliable elevation and orientation information, while object-based localization systems face challenges with frequent object changes requiring frequent model updates, increasing complexity and cost.
Innovation Solution
Utilize multiple imaging sensors to detect stationary objects in a vehicle's environment, compute their orientation relative to these objects, and combine this with geolocation data to achieve high-accuracy absolute positioning, including elevation and orientation, by employing trained machine learning models to identify and correlate objects across sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If satellite navigation data is used for vehicle localization, then positioning information can be obtained, but accuracy is insufficient and elevation/orientation data are unreliable
Solution Approach 1:
The patent introduces stationary objects (trees, poles, buildings, signs) as intermediary reference points between the vehicle and the positioning system. These objects serve as mediators that provide stable, verifiable geometric relationships for determining vehicle position, elevation, and orientation, replacing the unreliable satellite navigation data for these specific parameters.
Solution Approach 2:
The patent replaces the satellite-based electromagnetic positioning system with an object-based visual geometric system. By using imaging sensors to detect stationary objects and computing relative positioning based on object geometry and sensor orientation, the system substitutes the satellite navigation mechanism with a vision-based geometric computation approach.
2Measurement precision
If object-based localization is implemented, then localization accuracy can be improved, but frequent object changes require frequent model updates, increasing system complexity and cost
Solution Approach 1:
Instead of updating models to adapt to changing objects, the patent inverts the approach by selecting stationary objects that remain constant and using them as fixed reference points. The system adapts to object changes by relying on the stability of selected infrastructure objects rather than attempting to model all possible objects.
Solution Approach 2:
The patent changes the fundamental parameter from tracking all objects to tracking only stationary infrastructure objects with stable geolocations. By filtering for objects with immutable positions (trees, poles, buildings, signs), the system transforms the dynamic object recognition problem into a static reference point utilization problem, eliminating the need for frequent model updates.
3Measurement precision
If multiple imaging sensors are deployed to detect stationary objects and compute orientation, then high-accuracy absolute positioning including elevation and orientation can be achieved, but device complexity increases
Solution Approach 1:
The patent makes the imaging sensors multi-functional by using them simultaneously for detecting stationary objects, determining sensor orientation relative to those objects, and computing vehicle position and elevation. The same sensor array serves multiple localization functions, reducing the need for separate specialized sensors for each measurement type.
Solution Approach 2:
The patent merges the functions of object detection, orientation measurement, and position calculation into a unified processing framework. By combining the data from multiple imaging sensors and integrating it with stationary object geolocation information, the system achieves comprehensive absolute positioning (x, y, z, orientation) through a single integrated system rather than separate subsystems.
Data Source
AI summary
A method of self-localizing with respect to surrounding objects, comprising obtaining an approximated geolocation of the vehicle, retrieving mapping data comprising a geolocation of one or more stationary objects located in an area surrounding the approximated geolocation, receiving imagery data of a surrounding environment of the vehicle captured by a plurality of distinct imaging sensors deployed in the vehicle, applying one or more trained machine learning models to identify one or more of the stationary objects in the imagery data, computing a relative positioning of the vehicle with respect to one or more of the stationary objects based on an orientation of each of the plurality of imaging sensors with respect to the stationary object(s), computing an absolute positioning of the vehicle based on the relative positioning and the geolocation of the stationary object(s), and outputting the vehicle's absolute positioning.