Hybrid Neural Network Tracking for Sparse Radar Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Radar sensors in ADAS systems face challenges in accurately estimating the position, velocity, yaw angle, and classification of extended objects due to their sparse detection distribution, which complicates object tracking and classification, especially in poor visibility conditions.
Innovation Solution
A hybrid object tracking module utilizing a deep neural network model with modules such as a recurrent track feature abstractor, detection offset accumulator, and object regression heads to estimate bounding boxes and confidence levels based on sparse radar detections, incorporating modules like a recurrent object feature backbone and estimation confidence head to enhance object tracking accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If radar sensors are used for object detection in ADAS systems, then detection capability in poor visibility conditions is improved, but measurement precision of object parameters (position, velocity, yaw angle, classification) deteriorates due to sparse detection distribution
Solution Approach 1:
The system segments the object detection task into multiple specialized modules: a recurrent track feature abstractor for temporal feature extraction, an object regression head for parameter estimation, and an estimation confidence head for uncertainty quantification. Each module processes specific aspects of the sparse radar detections to collectively improve measurement precision while maintaining detection capability in poor visibility conditions.
2Measurement precision
If complex processing algorithms are applied to sparse radar detections to improve object tracking accuracy, then measurement precision improves, but processing power and memory requirements increase
Solution Approach 1:
The system performs preliminary action by pre-processing radar detections to extract track features and accumulate detection offsets before main processing. The recurrent track feature abstractor pre-computes temporal features from historical detections, and the detection offset accumulator pre-c prepares relative position information, reducing the computational burden on subsequent regression and confidence estimation stages.
3Measurement precision
If more radar detections are processed to improve bounding box estimation accuracy, then measurement precision improves, but processing time increases
Solution Approach 1:
The system extracts only the essential features from radar detections rather than processing all raw detection data. The recurrent track feature abstractor extracts temporal patterns and track characteristics, while the detection offset accumulator extracts relative position information. This selective extraction of critical features enables accurate bounding box estimation while minimizing processing time by ignoring redundant detection details.
Data Source
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.