Low-Level Sensor Fusion for Automated Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated driving and assistance systems face challenges in integrating data from multiple sensors with overlapping and non-overlapping fields of view, leading to inconsistencies in object detection and confidence levels, which affects decision-making and reaction times when objects move relative to the vehicle.
Innovation Solution
A computing system for low-level sensor fusion that temporally and spatially aligns raw measurement data from various sensors into an environmental coordinate field, allowing for the detection of objects and tracking their presence with improved confidence levels, and providing feedback for sensor calibration and operation adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If object-level integration is used to confirm object detection across sensors, then confidence level increases when multiple sensors detect the same object, but decision-making is delayed when sensors diverge in detection
Solution Approach 1:
The patent segments the sensor fusion process into two distinct levels: low-level fusion that continuously tracks objects across sensors using raw measurement data, and high-level fusion that performs object-level integration for confirmation. This segmentation allows the system to maintain continuous object presence information through low-level tracking while using high-level integration only when additional confirmation is needed, thereby reducing decision-making delays while preserving confidence level improvements.
Solution Approach 2:
The patent implements preliminary action by performing low-level sensor fusion in advance to establish continuous object presence and track objects across sensor fields before high-level object-level integration is needed. This preliminary tracking creates a foundation of confidence information that reduces the need for time-consuming sensor divergence decisions, as the system already has preliminary object presence data from the low-level fusion process.
2Adaptability or versatility
If each sensor performs separate object detection based on its own measurements, then sensor independence is maintained, but objects may be lost when moving between sensor fields of view
Solution Approach 1:
The patent merges sensor data at the low-level measurement stage by combining raw measurement data from multiple sensors into a unified environmental coordinate field. This merging creates a continuous object presence representation that tracks objects as they move between sensor fields of view, maintaining object detection reliability while preserving sensor independence at the data collection stage. The unified coordinate system allows seamless object tracking across sensor boundaries.
Solution Approach 2:
The patent introduces an intermediary environmental coordinate field that serves as a mediator between individual sensor measurement coordinate fields. This intermediary coordinate system receives raw measurement data from multiple sensors with different fields of view and integrates them into a unified representation, allowing objects to be tracked continuously as they move between sensor fields while maintaining the independence of individual sensor operations.
3Reliability
If multiple sensors are integrated to cover overlapping fields of view, then object detection reliability improves, but system complexity increases
Solution Approach 1:
The patent segments the sensor fusion complexity into two distinct processing levels: low-level fusion that handles raw measurement data alignment and integration in environmental coordinate fields, and high-level fusion that performs object-level integration. This segmentation allows the system to manage multiple sensors with overlapping fields of view by organizing the complexity into manageable processing stages, reducing overall system complexity while maintaining detection reliability.
Solution Approach 2:
The patent resolves sensor integration complexity by transforming the problem into a different dimensional space through the use of environmental coordinate fields. Instead of managing complexity in each sensor's native coordinate system, the patent projects all sensor measurements into a unified environmental coordinate field, simplifying the integration of multiple sensors with overlapping fields of view while maintaining detection reliability.
Data Source
AI summary
This application discloses a computing system to implement low-level sensor fusion in an assisted or automated driving system of a vehicle. The low-level sensor fusion can include receiving raw measurement data from sensors in the vehicle and temporally aligning the raw measurement data based on a time of capture. The low-level sensor fusion can include spatially aligning measurement coordinate fields of the sensors into an environmental coordinate field based, at least in part, on where the sensors are mounted in the vehicle, and then populating the environmental coordinate field with raw measurement data captured by the sensors based on the spatial alignment of the measurement coordinate fields to the environmental coordinate field. The low-level sensor fusion can detect at least one detection event or object based, at least in part, on the raw measurement data from multiple sensors as populated in the environmental coordinate field.