Traffic Sign Recognition via Element Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recognition systems, such as neural networks, require a large amount of information and calculation to recognize various targets, especially when dealing with a wide variety of traffic signs, leading to inefficient processing and high computational demands.
Innovation Solution
A recognition device that identifies the presence or absence of composition elements in input data using a trained identification device, such as a convolutional neural network, and calculates a score based on pre-defined relationships between element presence/absence and target types, reducing the computational load by focusing on common composition elements across similar indications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network is used to recognize various traffic signs, then recognition accuracy is improved, but the calculation amount and information processing requirements increase significantly
Solution Approach 1:
The patent segments the traffic sign recognition task into two independent stages: first, a classification unit identifies the type of traffic sign using a trained neural network; second, a recognition unit extracts specific information elements based on the classified type. This segmentation allows the neural network to be trained only for classification with a limited set of categories, rather than learning all possible information elements, thereby reducing the calculation amount while maintaining recognition accuracy.
Solution Approach 2:
The patent introduces an intermediary classification unit that acts as a mediator between the input image and the final recognition output. This classification unit categorizes traffic signs into predefined types, and each type has associated recognition rules that define which information elements to extract. This intermediary step reduces the complexity of the direct recognition task by breaking it down into classification followed by rule-based extraction.
2Adaptability or versatility
If the neural network is trained to recognize additional target types, then versatility is improved, but the number of parameters and training data requirements increase
Solution Approach 1:
The patent creates a universal recognition system where the classification unit is trained to identify multiple traffic sign types, and each type has a standardized set of recognition rules. The same recognition unit structure and rule extraction mechanism are applied universally across all traffic sign types, eliminating the need for separate recognition models for each type. This multi-functional approach allows the system to handle various target types while keeping the number of parameters manageable.
Solution Approach 2:
The patent changes the parameter representation by organizing recognition information as rules associated with each traffic sign type rather than as separate learnable parameters for each type. The recognition rules define which information elements to extract based on the classified type, transforming the problem from learning numerous parameters for each target type to applying predefined extraction rules, thereby reducing device complexity.
Data Source
AI summary
According to one embodiment, a recognition device includes a hardware processor. The hardware processor is configured to identify presence or absence of composition elements of a target from input observation data, acquire setting information indicative of a relationship between a type and composition elements, and identify a type of the target based on the setting information acquired and the presence or absence of the composition elements.