Autonomous Driving Obstacle Recognition Using Feature Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems face challenges in accurately recognizing obstacles, leading to potential erroneous recognition, which can interfere with the operation of autonomous vehicles.
Innovation Solution
The proposed autonomous driving system includes a recognition unit for identifying obstacles based on image data, a creation unit for generating a feature quantity distribution of past obstacle recognitions, and a judgement unit for comparing this distribution with the feature quantity of a recognized obstacle to determine if the recognition is erroneous.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If obstacle recognition is performed based on image data from external imaging devices, then obstacle detection capability is improved, but erroneous recognition occurs reducing reliability
Solution Approach 1:
The system establishes a feedback mechanism where recognition results are fed back into the system to create or update feature quantity distributions. These distributions represent statistical patterns of legitimate obstacles at different locations and times. The feedback loop enables continuous learning and refinement of recognition accuracy by comparing new recognition targets against historical patterns, thereby reducing erroneous recognition while maintaining high detection accuracy.
Solution Approach 2:
The system performs preliminary actions by pre-establishing feature quantity distributions based on historical recognition data before actual obstacle detection occurs. These pre-computed distributions serve as reference models that encode knowledge about typical obstacle characteristics at various locations and times. When a new obstacle is detected, the system can immediately compare it against these pre-prepared distributions to judge whether the recognition is legitimate, enabling rapid and accurate judgment without requiring real-time complex analysis.
2Measurement precision
If feature quantity distribution comparison is implemented to reduce erroneous recognition, then recognition accuracy is improved, but system complexity increases
Solution Approach 1:
The system creates simplified copies of obstacle characteristics in the form of feature quantity distributions. Instead of storing and processing complete image data or complex obstacle models, the system extracts and stores only the essential feature quantities (such as detection frequency, timing patterns, and spatial characteristics) that define legitimate obstacles. These feature quantity distributions serve as lightweight copies that can be rapidly compared against new recognition targets, achieving high recognition accuracy without requiring complex computational resources or large data storage capacities.
Data Source
AI summary
An autonomous driving system includes a recognition unit configured to recognize an obstacle based on image data obtained by imaging a predetermined region including a road on which an autonomous vehicle travels, the imaging being performed by an imaging device installed at a specific location in the external environment of the autonomous vehicle, a feature quantity distribution creation unit configured to create a feature quantity distribution expressing a distribution of features related to obstacles recognized in the past, and an erroneous recognition judgement unit configured to compare the created feature quantity distribution with a feature quantity of an obstacle recognized as an evaluation subject, to thereby judge whether the obstacle recognized as the evaluation subject is erroneously recognized.