Multilabel Traffic Sign Classification With Hierarchical Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification techniques for objects in autonomous vehicles, such as traffic signs, require large training datasets and high computational resources due to scalability limitations, especially in hierarchical and flat classification methods.
Innovation Solution
Implementing a multilabel hierarchical classification system using neural networks that leverage shared backbone layers, attribute and object classification subnetworks, and attribute and hierarchical classification layers to classify objects with fewer training data and computational resources, enabling zero-shot or few-shot learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If flat classification technique is used to classify traffic signs, then the neural network can directly classify traffic signs into specific categories, but the neural network requires a large training set with hundreds or thousands of training samples for each classification type
Solution Approach 1:
The patent segments the classification task into two parts: a hierarchical classification network that first identifies coarse-grained categories (e.g., traffic sign type), and a multilabel classification network that identifies fine-grained attributes (e.g., specific speed limit value). This segmentation allows each subnetwork to be trained on fewer samples compared to a single flat classification network that must distinguish all categories simultaneously.
Solution Approach 2:
The patent introduces a hierarchical dimension to the classification problem by organizing categories into a tree structure with parent-child relationships. This dimensional transformation allows the system to break down a high-dimensional classification problem into multiple lower-dimensional subproblems, reducing the training data requirement at each level.
2Adaptability or versatility
If hierarchical classification technique is used to classify traffic signs, then the system can classify images into parent and child classes, but each child classification still requires a large number of training samples
Solution Approach 1:
The patent creates a shared feature extraction backbone that serves both the hierarchical classification network and the multilabel classification network. This universal backbone is trained once on a combined dataset, and its learned features are reused by both classification tasks, eliminating the need to train separate backbones for each classification type and reducing overall training data requirements.
Solution Approach 2:
The patent merges the hierarchical classification task and the multilabel classification task into a unified framework where both networks share the same feature extraction backbone. This merging allows the system to leverage training samples efficiently across both classification objectives, reducing the total training data volume needed compared to training separate independent networks.
3Reliability
If conventional classification systems are trained to classify objects with no or little training data, then the system cannot achieve accurate classification, but training with large datasets increases computational cost and time
Solution Approach 1:
The patent performs preliminary action by pre-training a shared backbone network on a large but fixed dataset to learn generalizable features. Once the backbone is pre-trained, it can be fine-tuned or used directly for both hierarchical and multilabel classification tasks with minimal additional training data, avoiding the need to train from scratch for each classification problem and significantly reducing training time.
Data Source
AI summary
In various examples, multilabel hierarchical classification of objects for autonomous systems and applications is described herein. Systems and methods are disclosed that use one or more neural networks to classify objects, such as traffic signs, using multilabel classification and/or hierarchical classification. For instance, a multilabel subnetwork of the neural network(s) may classify an object based at least on one or more attributes associated with the object. As such, the output from the multilabel subnetwork may include at least a classification associated with the object and an attribute classification(s) associated with the object. A hierarchical subnetwork of the neural network(s) may also classify the object using one or more class labels, where a class label indicates another classification and/or a class group associated with the object. The systems and methods may then use the classification, the attribute classification(s), and/or the class label(s) to determine a final classification associated with the object.