Vehicle Sensor Data Filtering Using Compressed Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for collecting vehicle data, particularly for autonomous driving, face challenges in efficiently and intelligently gathering balanced data sets due to varying vehicle environments and the need to transmit, store, and process large quantities of data.
Innovation Solution
A method involving feature vector ascertainment, compression using hash functions, and local analysis units to filter and select data for transmission, reducing redundancy by comparing compressed representations across vehicles and a higher-level analysis unit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If all sensor data from multiple vehicles is transmitted to central analysis unit, then complete data collection is achieved, but data transmission volume and processing load become excessively large
Solution Approach 1:
The system segments data processing into three levels: local vehicle-level processing (sensor data to feature vectors), regional analysis unit processing (feature vectors to compressed representations), and central analysis unit processing (compressed representations to final datasets). This segmentation allows progressive filtering and compression at each level, reducing the volume of data transmitted to the central unit while maintaining representative samples of vehicle environments.
Solution Approach 2:
The system performs preliminary processing actions at local vehicles and regional analysis units before data reaches the central analysis unit. Feature vectors are extracted and compressed representations are generated in advance, allowing the central unit to receive pre-processed, reduced-volume data that still captures essential environmental characteristics for training purposes.
2Adaptability or versatility
If data from all vehicles is collected without filtering, then data diversity is maintained, but redundant data increases storage requirements
Solution Approach 1:
The system transforms sensor data through multiple parameter transformations: raw sensor data becomes feature vectors through extraction algorithms, which then become compressed representations through hashing and summarization. These parameter changes reduce data volume while preserving essential diversity characteristics needed for training neural networks on various vehicle environments.
Solution Approach 2:
Instead of transmitting all original sensor data, the system creates compressed copies (compressed representations) that capture the essential characteristics of vehicle environments. These compressed copies maintain diversity information while reducing storage requirements, allowing the central analysis unit to store and process only the most relevant data samples.
3Measurement precision
If feature vectors and compressed representations are generated for all data sets, then data quality is improved, but processing time increases
Solution Approach 1:
The processing task is segmented across multiple distributed components: local vehicles generate feature vectors independently, regional analysis units generate compressed representations from these vectors, and the central analysis unit processes these compressed data. This segmentation allows parallel processing and distributes computational load, reducing total processing time while maintaining data quality through multi-stage validation.
Solution Approach 2:
Feature vector extraction and compressed representation generation are performed as preliminary actions at distributed locations before central processing. This preliminary processing reduces the complexity of data at the source, allowing the central analysis unit to work with simplified data structures that require less processing time while maintaining sufficient accuracy for training purposes.
Data Source
AI summary
A method for collecting data from vehicles. Data sets are ascertained of at least one respective sensor of a plurality of vehicles. A respective feature vector is ascertained of the data sets. A respective compressed representation is ascertained of the feature vectors. The compressed representation is compared with a blacklist and/or a whitelist.