Vehicle LiDAR Object Classification Using Reflectivity Fingerprints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LIDAR systems for autonomous vehicles face limitations in accurately identifying and classifying objects due to inefficient and inaccurate distance-based object identification, particularly under varying environmental conditions, and are constrained by eye-safety regulations on maximum illumination power.
Innovation Solution
The system improves object identification and classification using LIDAR by detecting surface angles based on temporal distortions in reflection signals and employing reflectivity fingerprints, along with spatial and temporal relationships between different portions of an object's reflectivity to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LIDAR systems use distance information from reflected pulses to identify and classify objects, then the system can operate with basic sensing capabilities, but the identification and classification become error-prone and inaccurate
Solution Approach 1:
The patent applies local quality by analyzing specific local characteristics of object surfaces through reflectivity measurements at multiple discrete locations. Instead of relying on overall distance information, the system measures reflectivity at specific points across the object surface to create a detailed reflectivity profile that enables accurate identification and classification of objects even under varying environmental conditions.
Solution Approach 2:
The patent utilizes optical property variations (analogous to color changes) by measuring reflectivity characteristics across different portions of an object. The system detects variations in reflectivity intensity and patterns that serve as unique identifiers for different object types, enabling reliable classification based on optical signatures rather than distance alone.
2Length of stationary object
If LIDAR systems increase illumination power to detect far-away objects, then detection range improves, but the system must comply with eye-safety regulations that limit maximum illumination power
Solution Approach 1:
The patent applies parameter changes by transitioning from relying on illumination power to optimizing detection based on reflectivity measurement parameters. The system uses multiple discrete reflectivity measurements across different object portions, processing these measurements to identify objects at extended ranges without increasing illumination power, thereby maintaining eye-safety compliance while improving detection capability.
3Device complexity
If LIDAR systems use only distance information for object identification, then the system complexity remains low, but the efficiency and accuracy of object classification deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the object surface into multiple discrete measurement locations. The system performs reflectivity measurements at several distinct points across the object rather than using a single distance measurement, creating a segmented reflectivity profile that significantly improves classification efficiency and accuracy while maintaining manageable system complexity.
Solution Approach 2:
The patent transitions from one-dimensional distance measurement to multi-dimensional reflectivity characterization. By measuring reflectivity at multiple discrete locations and analyzing the spatial distribution and intensity variations, the system adds dimensional complexity to the data structure, enabling more efficient and accurate object classification without excessive system complexity.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances the accuracy and efficiency of object detection and classification in LIDAR systems, particularly for vehicles, by utilizing reflectivity fingerprints and temporal distortions to distinguish features like license plates, improving reliability in diverse conditions.
Implementation Method 1
receive, from at least one sensor, signals indicative of light reflected from a particular object in the field of view
Implementation Method 2
signals indicative of light reflected from a particular object in the field of view
Implementation Method 3
detect, based on time of flight in the received signals, portions of the particular object in the field of view that are similarly spaced from the light source
Data Source
AI summary
A method for classifying objects in a vehicle's surroundings includes receiving, on a pixel-by-pixel basis, a plurality of measurements associated with LIDAR detection results for a plurality of pixels of a field of view (FOV) of a LIDAR system configured to detect objects in at least part of the vehicle's surroundings, the measurements including at least one of: a presence indication, a surface angle, object surface physical composition, and a reflectivity level, receiving, on the pixel-by-pixel basis, at least one confidence level associated with each received measurement of the plurality of measurements, accessing at least one memory configured to store classification information for classifying a plurality of objects, and associating a plurality of pixels with a particular object based on the classification information and the received measurements with the at least one associated confidence level.