Feature-Based Data Reduction for Driver Assistance Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data reduction and compression methods for feature-based environmental information in driver assistance systems are insufficient, leading to memory capacity issues and high costs due to large data volumes, especially in embedded systems with limited resources.
Innovation Solution
A method that acquires and processes images from multiple positions using an optical sensor, extracts feature vectors, filters based on contrast information, and compares similarities to create a sparse feature matrix, which is then further reduced using compression techniques like run-length encoding and entropy coding, optimizing storage and transfer efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image processing methods are used to retain image information, then image quality is improved, but data volume reduction is insufficient for embedded memory constraints
Solution Approach 1:
The patent segments the environment detection data into multiple feature categories (geometric features, semantic features, spatial relationships) and processes each category separately. This allows selective retention of essential features while discarding redundant information, achieving data reduction without losing critical image information needed for driver assistance functions.
Solution Approach 2:
The patent extracts only the essential features from complete images - such as object boundaries, key semantic attributes, and critical spatial relationships - and stores these extracted features instead of the original images. This extraction process removes unnecessary pixel data while preserving the information needed for safety-critical decisions.
2Quantity of substance
If larger nonvolatile memory is used to store sensor data, then storage capacity is improved, but system cost increases
Solution Approach 1:
The patent extracts and stores only essential environmental features rather than complete sensor images, dramatically reducing the data volume that needs to be stored. This allows the use of smaller, more cost-effective nonvolatile memory components in embedded systems while still providing sufficient storage for the training and operation of driver assistance functions.
Solution Approach 2:
The patent transforms the data from high-resolution images to compact feature representations with significantly fewer parameters. This parameter reduction enables the use of smaller memory capacities, thereby reducing component costs and making the system more economically viable for mass production.
3Quantity of substance
If data reduction is applied to image information, then data volume is reduced, but relevant image information is discarded
Solution Approach 1:
The patent applies different processing quality levels to different regions and features of the environmental data. Critical features such as object boundaries, semantic attributes, and spatial relationships are preserved with high fidelity, while less critical information is reduced or discarded. This local quality differentiation ensures that relevant information is retained while achieving overall data reduction.
Solution Approach 2:
The patent performs preliminary feature extraction and identification during the training phase, identifying which features are most critical for specific driver assistance functions. This preliminary analysis allows the system to be configured to retain only the necessary information for each application, preventing loss of relevant data while maximizing reduction of unnecessary information.
4Speed
If higher data transfer rates are used, then data transmission speed is improved, but system cost and power consumption increase
Solution Approach 1:
The patent extracts and transfers only essential feature data rather than complete images, reducing the data volume that needs to be transmitted between processor and memory. This reduction in transfer volume allows the use of lower-performance, lower-cost data buses and interfaces while maintaining adequate transfer speeds for real-time operation.
Solution Approach 2:
The patent changes the data representation from high-bandwidth image formats to compact feature vector formats with fewer parameters. This parameter reduction enables the use of lower data transfer rates and lower-performance communication interfaces, reducing both system cost and power consumption while still providing sufficient bandwidth for the reduced data volume.
Data Source
AI summary
A method in a training phase includes acquiring a first image of an environment of a means of locomotion from a first position, using an optical sensor; acquiring a second image of the environment of the means of locomotion from a second position, differing from the first position, using the optical sensor; ascertaining features that represent the first image and ascertaining features that represent the second image using an algorithm for feature extraction; selecting those features of the first and the second image which do not meet a predefined rating criterion; and ascertaining significant similarities between the selected features of the first and the second image and storing references that represent the significant similarities. A method is performed in an execution phase that makes use of the stored features of the training phase.