Research Publications
Discover our latest research contributions to the scientific community. Our publications span journal articles, conference papers, and technical reports in radar technology and related fields.
CalibRefine: Deep Learning-Based Online Automatic Targetless LiDAR-Camera Calibration with Iterative and Attention-Driven Post-Refinement
L. Cheng, L. Guo, T. Zhang, T. Bang, A. Harris, M. Hajij, M. Sartipi, S. Cao
CalibRefine is a fully automatic, targetless, and online calibration framework that directly processes raw LiDAR point clouds and camera images. The approach features a Common Feature Discriminator, coarse homography-based calibration, iterative refinement, and attention-based refinement using Vision Transformer and cross-attention mechanisms to address non-planar distortions. Achieves high-precision calibration with minimal human input.
TransRAD: Retentive Vision Transformer for Enhanced Radar Object Detection
L. Cheng, S. Cao
TransRAD is a novel 3D radar object detection model that leverages the Retentive Vision Transformer (RMT) to learn features from radar Range-Azimuth-Doppler (RAD) data. The approach incorporates the Retentive Manhattan Self-Attention (MaSA) mechanism to align with spatial saliency characteristics of radar targets, and proposes Location-Aware NMS to mitigate duplicate bounding boxes in deep radar object detection.
mmWave Radar for Sit-to-Stand Analysis: A Comparative Study with Wearables and Kinect
S. Hu, P. Ackun, X. Zhang, S. Cao, J. Barton, M. G. Hector, M. J. Fain, N. Toosizadeh
This study presents a comparative analysis of mmWave radar for sit-to-stand movement assessment against established methods including wearable sensors and Kinect systems. The research demonstrates the potential of mmWave radar as a non-contact, privacy-preserving alternative for functional mobility assessment in clinical and home settings.
Deep Learning-Based Robust Multi-Object Tracking via Fusion of mmWave Radar and Camera Sensors
L. Cheng, A. Sengupta, S. Cao
A deep learning-based method integrating radar and camera data for multi-object tracking. Employs Bi-directional LSTM with tri-output mechanisms for robustness to noise and occlusion, combined with appearance feature model inspired by FaceNet for consistent tracking across frames.
Radar-Based Fall Detection: A Survey
S. Hu, S. Cao, N. Toosizadeh, J. Barton, M. G. Hector, M. J. Fain
A comprehensive survey of radar-based fall detection systems covering sensor technologies, signal processing techniques, machine learning approaches, and real-world deployment considerations. Emphasizes the effectiveness of Micro-Doppler, Range-Doppler, and Range-Doppler-Angle techniques in ensuring privacy and detecting falls through obstructions.
mmPose-FK: A Forward Kinematics Approach to Dynamic Skeletal Pose Estimation Using mmWave Radars
S. Hu, S. Cao, N. Toosizadeh, J. Barton, M. G. Hector, M. J. Fain
This paper presents mmPose-FK, a novel approach for human pose estimation using millimeter-wave (mmWave) radar combined with dynamic forward kinematics constraints. The method addresses challenges of low resolution, specularity, and noise artifacts common to mmWave radars by integrating FK into a deep learning model for more stable and accurate pose estimation.
Radar Range-Doppler Flow: A Radar Signal Processing Technique to Enhance Radar Target Classification
Q. Wen, S. Cao
This work presents Radar Range-Doppler Flow, a novel signal processing technique that enhances radar target classification capabilities. The method improves target discrimination and classification accuracy through advanced processing of range-Doppler data.
mmPose-NLP: A Natural Language Processing Approach to Precise Skeletal Pose Estimation using mmWave Radars
A. Sengupta, S. Cao
This work introduces mmPose-NLP, applying natural language processing concepts to radar-based pose estimation. The approach treats radar point clouds as sequential data, enabling more effective temporal modeling and improved pose tracking accuracy. First method to precisely estimate up to 25 skeletal key points using mmWave radar data alone.
Online Targetless Radar-Camera Extrinsic Calibration Based on the Common Features of Radar and Camera
L. Cheng, S. Cao
An online targetless calibration method leveraging deep learning to extract common features from raw radar Range-Doppler-Angle data and camera images. Uses RANSAC and adaptive variance Levenberg-Marquardt algorithm for robust calibration in uncontrolled environments.
3D Radar and Camera Co-Calibration: A Flexible and Accurate Method for Target-Based Extrinsic Calibration
L. Cheng, A. Sengupta, S. Cao
A flexible calibration method for 3D radar and camera fusion using a single corner reflector with PnP, RANSAC, and Levenberg-Marquardt optimization. The method does not require specially designed calibration environments and achieves accurate extrinsic calibration through iterative data collection.