Hybrid Neural Radar Tracking for Sparse Bounding Box Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Radar sensors in ADAS systems face challenges in accurately estimating the position, velocity, yaw angle, and classifying objects due to their sparse detection distribution, which makes it difficult to determine the size, orientation, and classification of detected objects, especially in poor visibility conditions.
Innovation Solution
A hybrid neural network-based object tracking system that utilizes a deep neural network model with modules like a recurrent track feature abstractor, detection offset accumulator, recurrent object feature backbone, object regression heads, and estimation confidence head to estimate bounding boxes and confidence levels based on sparse radar detections, incorporating radar detection modules and deep neural networks for enhanced object tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If radar sensors are used for object detection in ADAS systems, then the system can operate in poor visibility conditions, but the sparse detection distribution makes it difficult to accurately estimate object parameters such as position, velocity, yaw angle, and classification
Solution Approach 1:
The patent combines multiple radar detections into a unified object track by merging detection data points that belong to the same object. The deep neural network integrates sparse radar detections with centroid information from object tracking to generate comprehensive bounding box estimates, effectively combining fragmented detection data into accurate object parameter estimates.
Solution Approach 2:
The patent introduces a deep neural network as an intermediary between raw radar detections and final object parameter estimates. The DNN processes the sparse detection distribution and generates bounding box estimates, acting as a mediator that transforms incomplete radar data into accurate object parameters including position, velocity, yaw angle, and classification.
2Measurement precision
If a deep neural network model is used to estimate bounding boxes from sparse radar detections, then object tracking accuracy is improved, but processing power and memory requirements increase
Solution Approach 1:
The patent performs preliminary actions by generating object tracks and centroid information before feeding data to the deep neural network. The system pre-processes radar detections into organized detection distributions and prepares bounding box parameters in advance, reducing the computational burden on the DNN and optimizing processing efficiency.
3Productivity
If only peak radar detections are used to generate the sparse radar detection distribution, then processing speed is maximized, but the quantity of detection data is reduced
Solution Approach 1:
The patent extracts only the most significant radar detections (peak detections) from the full detection distribution. By selecting and extracting peak detections that contain the most relevant object information, the system reduces data quantity for faster processing while maintaining sufficient information for accurate bounding box estimation through the deep neural network.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A driver assistance system includes: a hybrid object tracking module comprising i) a radar detection module configured to receive a sparse radar detection distribution including radar detections based on a radar signal emitted from a host vehicle, ii) an object parameter determining module configured to generate an object track including centroid information for a detected object relative to the host vehicle, and iii) multiple modules implementing a deep neural network model and including neural networks, the deep neural network model configured to generate an estimate state of a bounding box and a confidence level of the estimated state of the bounding box based on the radar detections and the centroid information; and a driver assistance module configured to perform driver assistance operations based on the estimated state of the bounding box and the confidence level.