Sensor Data Quality Grading for Adaptive Fusion in Automated Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for sensor data fusion in automated driving systems fail to effectively utilize information about the reliability of different sensor modalities, leading to inefficiencies in mitigating false-positive and false-negative detections, particularly in scenarios requiring high accuracy like automatic emergency braking and collision avoidance.
Innovation Solution
A machine learning model, such as a neural network, is trained using data sets from multiple sensors of varying modalities to determine a quality grade for each sensor's data, leveraging ground truth objects and metrics to assess sensor reliability, enabling fusion methods like early, middle, and late fusion to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data fusion is optimized to avoid false-negative detections by assuming an object is real when perceived by at least one sensor modality, then collision avoidance capability is improved, but false-positive detection rate increases
Solution Approach 1:
The patent changes the parameter of object reality assessment from binary (real/not real) to a continuous quality grade scale. By introducing quality grades that reflect sensor reliability under different conditions, the system can dynamically adjust the threshold for assuming an object is real, thereby balancing false-negative and false-positive rates according to operational context.
Solution Approach 2:
The system dynamically adjusts the fusion strategy based on quality grades of sensor modalities. When sensor quality is high, the system can be more conservative; when quality degrades, it adapts by relying more on cross-modal verification. This dynamic adjustment resolves the contradiction by making the false-positive/false-negative trade-off controllable rather than fixed.
2Measurement precision
If sensor data fusion is optimized to avoid false-positive detections by assuming an object is real only when perceived by all sensor modalities, then false-positive detection rate is reduced, but false-negative detection rate increases
Solution Approach 1:
By introducing quality grades as a continuous parameter, the system replaces the all-or-nothing fusion approach with a graded assessment. Objects can be classified as real based on quality grade thresholds rather than requiring unanimous detection across all sensors, thus reducing false-negatives while maintaining false-positive control through the quality metric.
Solution Approach 2:
The fusion strategy becomes dynamic, adjusting the required level of sensor agreement based on quality grades. When quality grades indicate high reliability, the system can afford to be more stringent; when quality degrades, it lowers the bar for object confirmation, thereby maintaining detection completeness without sacrificing precision.
3Device complexity
If conventional sensor fusion methods are used without quality assessment, then system complexity is low, but sensor reliability information is not utilized leading to suboptimal detection accuracy
Solution Approach 1:
The system performs preliminary quality assessment of sensor modalities before fusion. By pre-evaluating sensor reliability under current conditions and assigning quality grades, the system prepares reliability information in advance that can be directly used during fusion without adding significant operational complexity, thus improving detection accuracy with minimal overhead.
Solution Approach 2:
The quality grade acts as an intermediary parameter between raw sensor data and fusion decision-making. This intermediary provides a compact representation of sensor reliability that can be easily integrated into existing fusion algorithms without requiring fundamental system redesign, thus bridging the gap between simplicity and accuracy.
Data Source
AI summary
A method for training a machine learning model for determining a quality grade of data sets from each of a plurality of sensors. The sensors are configured to generate surroundings representations. The method includes: providing data sets of each of the sensors from corresponding surroundings representations; providing attribute data of ground truth objects of the surroundings representations; determining a quality grade of the respective data set of each of the sensors using a metric, the metric comparing at least one variable, which is determined using the respective data set, with at least one attribute datum of at least one associated ground truth object of the surroundings representation; and training the machine learning model using the data sets of each of the sensors and the respectively assigned determined quality grades.