Template-Based Object Classification for New Sensor Classes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods for object classification in sensor data, such as those used in autonomous driving, are limited to recognizing known object classes and require extensive retraining for new classes, making them inefficient and unreliable.
Innovation Solution
A neural network trained with template data allows for fast and reliable classification of objects by using first and second template data to determine whether an object belongs to known or unknown classes, reducing the need for extensive retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If common machine learning methods are used for object classification, then the neural network can only recognize known object classes from training data, but extensive retraining is required for new object classes, making the system inefficient and inflexible
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network with template data representing various object classes before actual object classification is needed. This pre-training with diverse templates enables the network to quickly adapt to new object classes without extensive retraining, as the foundational recognition patterns are already established. The template data serves as advance preparation that reduces future adaptation time.
Solution Approach 2:
The patent introduces template data as an intermediary between the neural network and new object classes. These templates act as mediators that bridge the gap between existing training data and new objects, allowing the network to recognize new classes by comparing against templates rather than requiring complete retraining. The templates serve as reference intermediaries that facilitate quick adaptation.
2Reliability
If traditional neural networks are used for object classification, then the system requires extensive training data sets for each object class, but this increases the complexity and resource requirements of the system
Solution Approach 1:
The patent applies copying by using template data that represents object classes rather than requiring extensive actual training images for each class. The templates are simplified copies or representations that capture the essential features of object classes, enabling reliable classification with reduced data requirements. This copying approach maintains classification accuracy while reducing the burden of collecting and processing large training data sets.
3Adaptability or versatility
If the neural network is trained with multiple object classes, then the classification becomes more comprehensive, but the processing time and computational resources increase
Solution Approach 1:
The patent extracts and separates template data for different object classes, allowing the neural network to process and compare against specific templates rather than evaluating all possible classes simultaneously. This extraction approach enables comprehensive multi-class recognition while maintaining processing efficiency by organizing recognition into targeted comparisons with extracted template representations.
Data Source
AI summary
In a method for determining and classifying at least one object in a sensing area of a sensor, a two- or three-dimensional image of the sensing area is captured by means of the sensor. Sensor data corresponding to the image is generated. The object in the image is determined by means of the sensor data and by means of first template data of at least one first template object using a neural network. The neural network has been trained using the first template data. The first template data corresponds to an image of the first template object. At least one first object class is associated with the first template object. The object is classified by the neural network determining whether the object belongs to the first object class or not.