LiDAR Rain Point Filtering for Autonomous Vehicle Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
LiDAR sensors face degraded object recognition performance in rainy environments due to interference from raindrops, which are indistinguishable from object data.
Innovation Solution
A method to differentiate point data from a LiDAR sensor by determining specific criteria, such as intensity and spatial proximity, to identify raindrops and objects, and use these criteria to detect a rainy environment and improve object recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR sensor detects all point data in rainy environment, then complete object data is acquired, but raindrop point data interferes with object recognition
Solution Approach 1:
The patent segments point data into multiple layers based on distance from the LiDAR sensor. By dividing the detection space into near, mid, and far layers, the system can apply different processing strategies to different regions. Raindrops primarily affect the near layer, while object data is more prominent in mid and far layers, allowing selective filtering to improve object recognition accuracy.
Solution Approach 2:
The patent applies different intensity threshold criteria to different spatial layers. For near-layer point data, lower intensity thresholds are used to identify raindrops, while mid and far layers use higher thresholds to capture object data. This localized quality adjustment allows the system to distinguish raindrops from objects based on their spatial distribution and intensity characteristics.
2Measurement precision
If intensity threshold is set low to detect raindrops, then raindrop detection improves, but object data may be misclassified as raindrops
Solution Approach 1:
The patent divides point data into multiple layers based on distance, allowing different intensity threshold settings for each layer. Near-layer data uses lower thresholds for raindrop detection, while mid and far layers use higher thresholds to ensure object data is not misclassified. This segmentation resolves the contradiction by applying context-appropriate thresholds to different spatial regions.
Solution Approach 2:
The system adjusts intensity threshold criteria locally for different spatial layers. Near-layer point data is evaluated with lenient thresholds to capture raindrops, while mid and far layer data uses strict thresholds to maintain object recognition reliability. This local quality differentiation allows simultaneous optimization of raindrop detection and object recognition.
3Measurement precision
If all point data is processed for object recognition, then comprehensive detection is achieved, but processing time increases
Solution Approach 1:
The patent segments point data into multiple layers and processes each layer with appropriate filtering criteria. By identifying and filtering raindrop point data in the near layer before object recognition processing, the system reduces the volume of data requiring comprehensive object analysis, thereby decreasing processing time while maintaining detection completeness for mid and far layers.
Solution Approach 2:
The system extracts and separates raindrop point data from the overall point cloud by applying layer-specific intensity thresholds. By removing raindrop data from the near layer before object recognition processing, the system eliminates unnecessary computational overhead while preserving all relevant object data in mid and far layers, achieving time efficiency without sacrificing detection completeness.
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 object recognition accuracy in rainy conditions by distinguishing raindrops from object data, allowing for improved detection and tracking of objects.
Implementation Method 1
laser transmission pulses are emitted through a 1-st layer to an N-th Layer of the LiDAR sensor
Implementation Method 2
acquiring all the point data from laser reflected pulses, each of which is detected in response to each of the laser transmission pulses
Data Source
AI summary
Methods disclosed herein provide rainy environmental information by, on condition that a detection area is determined as an area of within a specific distance from a LiDAR sensor mounted on an autonomous vehicle, and laser transmission pulses are emitted through a 1-st layer to an N-th Layer of the LiDAR sensor, a computing device acquires all the point data from laser reflected pulses, within the detection area. Additionally, the computing device accumulates specific point data, among all the point data, which satisfy a specific criterion related to a rainy environment over a period of time, and detects the rainy environment by referring to the accumulated specific point data.