Semantic Segmentation Localization for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately determining their location in environments where GPS signals are unreliable, such as tunnels, and rely on error-prone systems like odometers and IMUs, which can lead to compounded errors over time.
Innovation Solution
The use of a semantic segmentation localization component that leverages semantically segmented images to detect frequently occurring objects like lane markers, using machine-learned models to extract features and combine them with map data for accurate localization, reducing reliance on GPS and other systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS and traditional localization systems (odometers, IMUs) are used for vehicle localization, then the system can provide location information, but the accuracy deteriorates in environments like tunnels where GPS signals are unreliable and errors compound over time
Solution Approach 1:
The patent introduces an intermediary localization system that uses semantically segmented images and machine-learned models to detect recurring environmental features (lane markers, curbs, sidewalks). This intermediary system bridges the gap when GPS fails, providing accurate localization by matching detected features with pre-stored semantic maps of the environment, thus resolving the contradiction between reliability and measurement precision.
Solution Approach 2:
The patent replaces reliance on mechanical localization systems (odometers, IMUs) that accumulate errors over time with a vision-based semantic segmentation system. By using machine-learned models to process images and detect environmental features, the system substitutes mechanical measurement methods with optical and computational approaches, achieving superior accuracy without GPS while maintaining continuous operation.
2Measurement precision
If semantic segmentation and machine-learned models are used to detect recurring objects, then localization accuracy improves, but processing resources increase
Solution Approach 1:
The patent extracts and processes only the most critical and recurring environmental features (lane markers, curbs, sidewalks) rather than analyzing all objects in the scene. By focusing computational resources on these specific, high-value features that provide the most localization information, the system achieves high accuracy while minimizing processing overhead and energy consumption.
Solution Approach 2:
The patent performs preliminary actions by pre-storing semantic maps of the environment and pre-training machine-learned models to recognize recurring features. This offline preparation reduces the computational burden during real-time operation, as the system only needs to match detected features against pre-processed data, significantly lowering on-the-fly processing requirements while maintaining high localization accuracy.
Data Source
AI summary
Techniques are discussed for determining a location of a vehicle in an environment using a feature corresponding to a portion of an image representing an object in the environment which is associated with a frequently occurring object classification. For example, an image may be received and semantically segmented to associate pixels of the image with a label representing an object of an object type (e.g., extracting only those portions of the image which represent lane boundary markings). Features may then be extracted, or otherwise determined, which are limited to those portions of the image. In some examples, map data indicating a previously mapped location of a corresponding portion of the object may be used to determine a difference. The difference (or sum of differences for multiple observations) are then used to localize the vehicle with respect to the map.