Multi-Radar Object Detection Using Cross-Potential Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately detecting and identifying objects in adverse weather conditions due to the limitations of light-based sensors like LiDAR, which fail in fog, dust, and snow, while radar systems suffer from specular reflections, sparsity, and noise in point clouds, leading to poor performance in generating accurate 3D bounding boxes.
Innovation Solution
A multi-radar system with spatially separated radars is used to generate cross potential point clouds, leveraging spatial diversity and deep learning architectures to filter noise and enhance signal quality, enabling accurate 3D bounding box estimation by combining data from multiple radars and applying noise filtering algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If light-based sensors (LiDAR) are used for object detection, then detection accuracy in good weather is improved, but reliability in adverse weather conditions deteriorates
Solution Approach 1:
The patent transitions from light-based sensing parameters to radio wave-based sensing parameters by using radar sensors. This parameter change allows the system to maintain detection capabilities in adverse weather conditions where light-based sensors fail, as radio waves can penetrate fog, rain, and snow that block light
Solution Approach 2:
The patent replaces the optical/mechanical LiDAR system with an electromagnetic radar-based system. This substitution fundamentally changes the sensing mechanism from light reflection measurement to radio wave reflection measurement, enabling operation in weather conditions that block visible light
2Device complexity
If single radar sensor is used for object detection, then device complexity is reduced, but measurement precision of 3D bounding boxes deteriorates
Solution Approach 1:
The patent combines data from multiple radar sensors positioned at different locations on the vehicle to generate a unified 3D bounding box estimation. By merging the point cloud data and detection results from multiple radars, the system achieves superior measurement precision that exceeds any single radar sensor
Solution Approach 2:
The patent divides the detection task among multiple radar sensors positioned at different locations (front, rear, sides), with each sensor handling a specific spatial sector. This segmentation allows each sensor to focus on its local region while collectively providing comprehensive 3D coverage
3Reliability
If radar sensors are used for all-weather detection, then reliability in adverse weather is improved, but measurement precision deteriorates due to specular reflections and noise
Solution Approach 1:
The patent converts the harmful effects of specular reflections and noise into beneficial information by using deep learning architectures that are specifically trained to recognize and filter these artifacts. The system learns to distinguish between genuine object reflections and spurious reflections from smooth surfaces, turning a previously problematic feature into a detectable pattern
Solution Approach 2:
The patent introduces deep learning algorithms as an intermediary processing layer between the raw radar signals and the final 3D bounding box output. This intermediary layer filters noise, corrects specular reflection artifacts, and enhances the quality of point cloud data before it is used for object detection
4Measurement precision
If multiple radar sensors are deployed for improved detection accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent designs the multi-radar system with a unified processing architecture that handles data from multiple sensors through a single deep learning model. This universal approach allows the same processing pipeline to handle data from any number of radar sensors, reducing the complexity that would otherwise arise from needing separate processing paths for each sensor
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
The system achieves improved accuracy in localizing object bounding boxes with median errors of less than 37 cm in center localization and 25 cm in dimension estimation, and a 48% performance improvement in mean-average precision compared to single-radar systems, while maintaining real-time inference capabilities.
Implementation Method 1
receiving one or more signals reflected by one or more second objects, the signals being received by one or more radar sensors
Data Source
AI summary
A method, a system, and a computer program product for detecting one or more objects. One or more signals reflected by one or more second objects is received, where the signals are received by one or more radar sensors positioned on one or more first objects. Based on the one or more received signals, one or more representations are generated. One or more portions of the generated representations correspond to the one or more received signals. One or more virtual enclosures encompassing the one or more second objects are generated using the one or more representations. A presence of the one or more second objects is detected using the generated one or more virtual enclosures.