Object Classification Disambiguation via Expectation Counts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous and semi-autonomous vehicles face challenges in accurately classifying traffic objects, leading to ambiguous data and potential misoperation due to incorrect object classification by onboard sensors, which affects navigation and map updates.
Innovation Solution
A method that disambiguates object classifications by considering the expectation count of nearby objects of the same type within a predefined distance, using information from multiple vehicles to determine an accurate classification and update mapping models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If object classification is performed using onboard sensors in autonomous vehicles, then navigation capability is enabled, but classification accuracy deteriorates leading to ambiguous data and potential misoperation
Solution Approach 1:
The patent combines object classification data from multiple vehicles (crowdsourcing) to resolve ambiguous classifications. When one vehicle's sensor data is uncertain, the system aggregates classifications from other vehicles observing the same object, merging multiple data sources to achieve higher classification accuracy and reliability.
Solution Approach 2:
The system implements feedback by using expectation counts (how many objects of a certain type are expected at a location) to validate and correct individual vehicle classifications. If the aggregated classification contradicts the expectation count, the system adjusts the classification to align with the expected pattern, ensuring consistent and reliable navigation data.
2Quantity of substance
If multiple object classifications are collected from multiple vehicles, then data completeness is improved, but data ambiguity increases making it difficult to determine the correct classification
Solution Approach 1:
The patent introduces expectation count as an additional parameter to evaluate object classifications. Instead of relying solely on raw classification counts, the system compares actual classifications against expected classifications based on location and object type patterns, using this parameter to resolve ambiguities and determine the most accurate classification.
Solution Approach 2:
The expectation count acts as an intermediary that mediates between multiple vehicle classifications and the final determined classification. When classifications from multiple vehicles conflict, the expectation count serves as a reference standard to resolve the conflict and establish the correct classification, preventing information loss.
3Measurement precision
If expectation count is used to disambiguate object classifications, then classification accuracy is improved, but computational complexity increases
Solution Approach 1:
The system applies partial action by using expectation counts selectively - only when classification ambiguity is detected. Rather than computing expectation counts for every single object classification, the system activates this more complex computation only when needed to resolve ambiguities, balancing accuracy improvement with computational efficiency.
Data Source
AI summary
Methods described herein relate to disambiguating objects for a particular location. Methods may include receiving, from a vehicle, an indication of an object associated with an object classification for the object in view of the vehicle. The object classification is further associated with an object type. The method further includes receiving, from the vehicle, information pertaining to one or more other objects of the same object type within a predefined distance of the object; determining an expectation count for the object type based at least in part on a count of the object and the one or more objects of the same object type for the particular location; and disambiguating one or more object classifications from a plurality of object classifications based at least in part on the expectation count. A corresponding apparatus and computer program product are also provided.