Granularity-flexible LiDAR Object Detection via Heuristic Point-Cloud Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face inefficiencies in object detection due to large amounts of LiDAR data, which can overwhelm computational resources, and existing methods do not effectively filter or scale this data for optimal granularity levels.
Innovation Solution
The method involves obtaining LiDAR data, aggregating point-clouds, applying heuristic filters to reduce data points based on physical constraints, and iteratively determining an operational granularity level to generate an array of existence-based objects, optimizing data processing for accurate and efficient object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR data is collected at high resolution to improve object detection accuracy, then measurement precision is improved, but the computational load increases and processing efficiency deteriorates
Solution Approach 1:
The point cloud data is divided into multiple local regions or clusters based on spatial proximity. Each region is processed independently to identify local structures and objects, reducing the overall computational complexity while preserving detection accuracy. This segmentation allows parallel processing and reduces the memory burden of handling entire point clouds at once.
Solution Approach 2:
The patent extracts and removes redundant or irrelevant points from the point cloud data before processing. This includes filtering out points that do not contribute to object detection, such as ground points or points from static background elements. By extracting only the essential features and points, the system maintains high detection accuracy while significantly reducing computational load.
2Reliability
If all LiDAR data points are processed to ensure complete object detection, then detection reliability is improved, but the computational resources required increase
Solution Approach 1:
The system performs preliminary filtering and preprocessing operations on the LiDAR data before main object detection algorithms are applied. This includes initial noise filtering, outlier removal, and coarse segmentation that eliminates obviously irrelevant points early in the pipeline. By performing these actions preliminarily, the system ensures reliable detection of actual objects while avoiding wasteful computation on irrelevant data points.
Solution Approach 2:
Different processing strategies are applied to different regions of the point cloud based on their local characteristics. High-density regions containing potential objects receive more intensive processing, while low-density or uniform regions receive lighter processing. This local quality approach ensures that computational resources are focused where they are most needed for reliable detection, rather than uniformly processing all points.
3Productivity
If data filtering is applied to reduce computational load, then processing efficiency is improved, but the risk of removing relevant data points increases
Solution Approach 1:
The filtering operations use dynamically adjusted parameters based on the local characteristics of the point cloud data. For example, density thresholds, distance thresholds, and noise filters are adapted to the specific region being processed. This allows aggressive filtering in safe regions while maintaining conservative filtering near potential objects, thus improving efficiency without compromising detection reliability.
Solution Approach 2:
The system incorporates feedback mechanisms where the results of preliminary filtering are used to adjust subsequent processing parameters. If filtering removes points that could have been object candidates, the system learns from this and adjusts its filtering criteria to be more conservative in similar situations. This feedback loop ensures that efficiency improvements do not come at the cost of detection reliability.
Data Source
AI summary
A method may include obtaining sensor data from one or more LiDAR units and determining a point-cloud corresponding to the sensor data obtained from each respective LiDAR unit. The method may include aggregating the point-clouds as an aggregated point-cloud. A number of data points included in the aggregated point-cloud may be decreased by filtering out one or more of the data points according to one or more heuristic rules to generate a reduced point-cloud. The method may include determining an operational granularity level for the reduced point-cloud. An array of existence-based objects may be generated based on the reduced point-cloud and the operational granularity level.