Deep Learning Network for Vision-Based Radar Point Cloud Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing radar systems for vehicle safety rely on expensive sensors and can produce inaccurate data, while machine learning techniques require large amounts of training data, which is often difficult to obtain.
Innovation Solution
A deep learning network is trained using vision sensor data to produce estimated sensor point cloud distributions, leveraging available vision sensor data to minimize the need for radar or other sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If radar sensors are used to improve vehicle safety, then detection accuracy is improved, but system cost increases
Solution Approach 1:
The patent creates a virtual copy of radar sensor data by training a deep learning network to generate synthetic radar point clouds from vision sensor images. The trained network produces estimated sensor point cloud distributions that replicate radar detection outputs, enabling radar-like functionality without physical radar sensors. This copying approach resolves the contradiction by achieving detection accuracy through software-generated data rather than expensive hardware.
2Measurement precision
If machine learning techniques are used to improve radar data, then detection accuracy is improved, but training data requirements increase
Solution Approach 1:
The patent uses vision sensor data, which is abundant, inexpensive, and easily obtained, as training data instead of requiring scarce and difficult-to-obtain radar data. The deep learning network is trained on readily available vision images and their corresponding ground truth annotations, eliminating the need for large volumes of expensive radar training data. This approach resolves the contradiction by substituting cheap, abundant vision data for expensive, scarce radar training data.
Data Source
AI summary
Techniques for using machine learning to produce sensor data from vision sensor data are disclosed. By using a limited amount of sensor data together with vision sensor data, a deep learning network can be trained to produce estimated sensor point cloud distributions from, e.g., vision sensor data alone. Using a deep learning network trained in this way, vehicles with limited or no other sensor functionality can be equipped with a camera to produce estimated sensor point cloud distributions. The estimated sensor point cloud distributions can then be used to improve vehicle safety through vehicle controls or driver notifications and/or to produce enhanced sensor data.