Multi-Channel LIDAR Matrix for Object Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately detecting and classifying objects proximate to them, particularly in distinguishing smaller objects from larger ones when they are close, which can lead to segmentation errors affecting motion planning and safety.
Innovation Solution
A computer-implemented method using a multi-channel data matrix generated from LIDAR data, including Range, Height, Intensity, Absence of LIDAR Return, and LIDAR Background Subtraction Foreground channels, is input into a machine-learned model, such as a convolutional neural network, to improve object detection and classification, enabling more accurate segmentation and tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional LIDAR-based object detection is used, then the system can detect objects in the environment, but it fails to accurately distinguish smaller objects from larger ones when they are close, leading to segmentation errors
Solution Approach 1:
The patent applies segmentation by dividing the LIDAR data processing into multiple distinct channels (Range, Height, Intensity, Absence of LIDAR Return, and LIDAR Background Subtraction foreground channels). Each channel processes specific aspects of the data independently, allowing the system to better distinguish between overlapping objects of different sizes and improve segmentation accuracy.
Solution Approach 2:
The patent transforms the LIDAR data into a multi-dimensional representation by creating a data matrix with multiple channels. This dimensional expansion allows the machine-learned model to analyze objects from multiple perspectives simultaneously, improving the ability to distinguish smaller objects from larger ones in close proximity.
2Measurement precision
If multi-channel data matrix with machine-learned model is used, then object detection accuracy and segmentation precision are improved, but computational complexity and processing requirements increase
Solution Approach 1:
The patent applies preliminary action by pre-processing the LIDAR data into a structured multi-channel data matrix before feeding it to the machine-learned model. The data is organized into specific channels (Range, Height, Intensity, etc.) in advance, which simplifies the model's processing task and reduces computational complexity during inference.
Solution Approach 2:
The patent changes the parameters of the data representation by transforming raw LIDAR data into a multi-channel data matrix with specific characteristics (Range, Height, Intensity, Absence of LIDAR Return, LIDAR Background Subtraction foreground). This parameter transformation enables the use of efficient machine-learned models while maintaining high detection accuracy.
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
This approach enhances object detection accuracy, reduces segmentation errors, and improves tracking and classification, leading to safer and more efficient autonomous vehicle operations by providing cleaner segmented objects and improved motion planning.
Implementation Method 1
LIDAR data from one or more LIDAR sensors configured to transmit ranging signals
Implementation Method 2
LIDAR sensors configured to transmit ranging signals
Data Source
AI summary
Systems and methods for detecting and classifying objects that are proximate to an autonomous vehicle can include receiving, by one or more computing devices, LIDAR data from one or more LIDAR sensors configured to transmit ranging signals relative to an autonomous vehicle, generating, by the one or more computing devices, a data matrix comprising a plurality of data channels based at least in part on the LIDAR data, and inputting the data matrix to a machine-learned model. A class prediction for each of one or more different portions of the data matrix and/or a properties estimation associated with each class prediction generated for the data matrix can be received as an output of the machine-learned model. One or more object segments can be generated based at least in part on the class predictions and properties estimations. The one or more object segments can be provided to an object classification and tracking application.