Radar Object Recognition System Using Non-Maximum Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radar object recognition systems in autonomous driving technologies suffer from decreased recognition accuracy due to the lack of processing and optimization of radar data, leading to inefficiencies in object detection.
Innovation Solution
A radar object recognition system that includes a processor for generating radar images through normalization, target enhancement, and Cartesian coordinate conversion, and utilizes a deep learning object recognition model with non-maximum suppression to eliminate overlapping bounding boxes and improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If radar data is directly input into object recognition model without processing, then processing time is reduced, but recognition accuracy decreases
Solution Approach 1:
The patent applies preliminary action by performing radar image generation, normalization, and target enhancement processing on radar data before inputting it into the object recognition model. This preprocessing prepares the data in advance to improve recognition accuracy while managing complexity through systematic processing steps.
Solution Approach 2:
The patent introduces radar image generation as an intermediary process between raw radar data and the object recognition model. This intermediary converts radar data into image format, enabling the model to process and recognize objects more accurately while maintaining manageable system complexity.
2Reliability
If multiple bounding boxes are generated for object detection, then detection coverage is improved, but overlapping errors increase
Solution Approach 1:
The patent applies taking out by extracting and removing overlapping bounding boxes from the detection results. The post-processing step identifies and eliminates redundant overlapping boxes, keeping only the most accurate detection results to maintain both coverage and precision.
Solution Approach 2:
The patent applies parameter changes by adjusting bounding box parameters through post-processing operations. It modifies confidence thresholds and applies non-maximum suppression to optimize the bounding box parameters, resolving overlapping errors while maintaining detection coverage.
3Measurement precision
If radar data map is converted to two-dimensional image, then object recognition is improved, but processing steps increase
Solution Approach 1:
The patent applies preliminary action by pre-converting radar data maps into two-dimensional radar images with enhanced target features before recognition. This preparation in advance improves object recognition accuracy by presenting optimized visual data to the recognition model.
Solution Approach 2:
The patent introduces two-dimensional radar image generation as an intermediary transformation between radar data maps and the object recognition model. This intermediary conversion enhances recognition capabilities by translating radar data into a more recognizable image format.
Data Source
AI summary
The present disclosure provides a radar object recognition method, which includes steps as follows. The radar image generation is performed on radar data to generate a radar image; the radar image is inputted into an object recognition model, so that the object recognition model outputs a recognition result; the post-process is performed on the recognition result to eliminate recognition errors from the recognition result.