CLEAR: Counting and Location Estimate using Ambient Radio signals

2015-2020

This project implemented novel high-resolution Radio-Frequency (RF) imaging in the indoor environment to monitor untagged people or objects.

ARPA-E summit occupant localization demonstration with a 1/6 scaled model. The screen shows detected locations of two figures in the model.

Project details and code can be found here: CLEAR

Implemented novel sparsity-based OMP and FISTA reconstruction algorithms for high-resolution RF imaging, using untagged-object backscattered phase from ambient low-cost passive UHF RFID tags.
Developed simulation study in CST Microwave Studio to compare relative tag-receiver placement and algorithm performance.
Designed a background-subtraction calibration algorithm with improved noise-cancellation, detecting one occupant presence with 100% accuracy and low median error of 0.36 m in a 4 m × 4 m room.
Developed ensemble algorithms to improve the performance and reduce parameter sensitivity for inverse methods with limited bandwidth and spatial diversity.
Implemented convex optimization and level-set approaches to extract object shape and size from the pixel reflectivity of the imaging domain.
Developed an optimal frequency selection algorithm for a broad bandwidth, multi-frequency setup to generate improved Fourier-reconstructed image based on K-space sampling.
Tested super-resolution imaging based Capon and maximum entropy algorithms, which provided improved performance over matched filtering for shape estimation with increased computational and time complexity.

True and detected occupant locations using 3D image reconstruction in a real-life setup.