ADAS Data Recording With Real-Time Scenario Metadata Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for data collection and validation of Advanced Driving Assistance Systems (ADAS) are inefficient, requiring extensive driving distances, significant storage space, delayed data analysis, and inadequate selection of relevant data for training AI algorithms, with event-based recording being unsuitable for early-stage ADAS development.
Innovation Solution
A system that utilizes sensor modules on vehicles to acquire data and generate metadata in real-time, allowing for on-the-fly data selection and processing, including classification of driving environments and scenarios, to optimize data collection and reduce redundant recording.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If continuous data recording is performed to meet predefined road scenario requirements, then sufficient training data is collected, but storage space requirements increase significantly
Solution Approach 1:
The system performs preliminary classification of driving scenarios and metadata generation during the data recording phase, before data is transferred to the data center. This allows redundant data to be identified and excluded from storage early in the process, preventing unnecessary storage consumption while ensuring sufficient training data is captured.
Solution Approach 2:
The patent applies different processing qualities to different data segments based on their relevance. Critical driving scenarios are fully recorded and processed, while redundant portions are minimized or excluded. This selective quality approach ensures training data sufficiency without uniform high-storage consumption across all recorded data.
2Reliability
If data is recorded and stored on-vehicle for later analysis, then data quality is preserved, but data analysis is delayed by weeks
Solution Approach 1:
The system performs preliminary processing of data quality assessment, metadata generation, and scenario classification during the recording phase itself. This preliminary action enables immediate evaluation of data quality and relevance, allowing rapid identification of useful training data without waiting for centralized analysis, thus reducing the weeks-long delay while maintaining quality standards.
Solution Approach 2:
The system implements real-time feedback mechanisms where recorded data is immediately processed and evaluated for quality and relevance. This feedback loop allows the system to assess data adequacy during the recording campaign and make real-time adjustments, eliminating the delayed feedback that previously required weeks of post-processing analysis.
3Volume of stationary object
If event-based recording is used to reduce storage requirements, then storage space is optimized, but the ADAS ECU must be mature enough to judge which situations to record
Solution Approach 1:
The patent introduces an intermediary processing layer that performs scenario classification and data relevance assessment. This intermediary component bridges the gap between raw sensor data and storage decisions, enabling intelligent selection of recording content without requiring the ADAS ECU itself to be fully mature. The intermediary handles the complexity of judgment, allowing the ECU to focus on core driving assistance functions.
4Quantity of substance
If hundreds of thousands of kilometers are recorded to meet strict requirements, then comprehensive coverage is achieved, but the recording campaign duration and cost increase
Solution Approach 1:
The system performs preliminary scenario classification and data adequacy assessment during the recording campaign. This preliminary action enables real-time determination of when sufficient coverage has been achieved for each scenario type, allowing the campaign to be terminated early once requirements are met, rather than continuing for fixed predetermined distances or times.
Solution Approach 2:
The patent dynamically adjusts recording parameters based on real-time assessment of data coverage and scenario diversity. When sufficient coverage is detected for certain scenario types, the system can reduce recording intensity or focus resources on underrepresented scenarios, optimizing the balance between comprehensive coverage and campaign duration.
Data Source
Figure 1
Figure 2
Figure 3(a)~3(f)
AI summary
A system (1) for assessing progress of a data-recording campaign performed to collect sensor data (S1) recorded by a sensor data recorder (30) mounted on a vehicle (5), the system comprising a metadata-generating apparatus (40) arranged to process data (S1, S2) acquired by a sensor module (20) to generate metadata (M) for the sensor data, the metadata comprising classifications of an attribute of a driving environment the vehicle was in during the acquisition of the sensor data into respective classes of a predefined set of classes of the attribute, and transmit the metadata to a remote metadata-processing apparatus (60), which is arranged to determine whether the metadata comprises at least a predetermined number of classifications in a predefined class of the set of predefined classes and, based on the determination, generate an indicator (T) for use in the data-recording campaign.