Radar Perception Grid Filtering for Efficient Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional radar systems for autonomous vehicles are computationally intensive and lack efficiency and robustness in generating radar perception grids.
Innovation Solution
A perception grid-based solution that filters radar scan data to remove detections outside a reachable region, false alarms, and low radar cross-section detections, and generates particles using a range-azimuth framework, with bounding boxes and least-square regression for refinement, enabling efficient and accurate object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional radar systems process all radar scan data to generate perception grids, then object detection coverage is comprehensive, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent extracts and removes false alarm detections and low radar cross-section detections from the radar scan data before processing. By filtering out these unnecessary detections that contribute to computational complexity without adding valuable detection coverage, the system reduces processing burden while maintaining reliable object detection for valid targets.
Solution Approach 2:
The patent segments the radar detection data into different categories: valid detections, false alarms, and low radar cross-section detections. This segmentation allows the system to process only the meaningful data segments (valid detections) while discarding or separately handling the non-essential segments, thereby reducing overall computational complexity while preserving detection reliability.
2Loss of information
If conventional radar systems process all radar scan data including false alarms and low RCS detections, then no data is lost, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary filtering actions before the main processing stage. By预先 removing false alarm detections and low radar cross-section detections from the dataset before generating perception grids, the system eliminates unnecessary processing steps that would waste computational time, while ensuring that all valid detection information is preserved for accurate object detection.
3Reliability
If radar systems use traditional track-based solutions for object detection, then detection methodology is conventional and well-established, but computational efficiency and robustness are insufficient
Solution Approach 1:
The patent inverts the conventional approach by not relying on traditional track-based solutions that require extensive processing of detection sequences. Instead, it directly processes radar scan data with preliminary filtering to generate perception grids, reversing the conventional workflow to achieve both robustness through proper data filtering and efficiency by avoiding computationally intensive tracking algorithms.
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 solution provides improved computational efficiency and robustness in generating radar perception grids, enhancing the accuracy and speed of object detection for autonomous vehicles.
Implementation Method 1
The received signal provides information about the object's location and speed. For example, if an object is moving either toward or away from the radar system, the received signal will have a slightly different frequency than the frequency of the emitted signal due to the Doppler effect.
Data Source
AI summary
This document discloses system, method, and computer program product embodiments for operating a radar system. For example, the method includes: obtaining radar scan data for a first scan performed by a first radar device and radar scan data for a second scan performed by the first radar device or a second device; identifying a detection pair comprising a first detection in the radar scan data for the first scan and a second detection in the radar scan data for the second scan; obtaining a 2D initial particle velocity using positions of first and second detections of the detection pair; using the 2D initial particle velocity to obtain a projected range rate for the second detection of the detection pair; and creating a particle when the projected range rate matches a range rate associated with the second detection of the detection pair by a certain amount.