Reza Baharani

Charlotte, NC

I build machine learning systems that run on the edge — from clinical health sensing on Apple devices to scalable video intelligence on embedded GPUs. My work spans the full ML pipeline: biomedical signal processing, deep learning architecture design, on-device optimization (CoreML, TensorFlow Lite), and real-world clinical validation.

Currently, I’m a Senior AI Engineer at ForesightCares, where I develop real-time 3D human pose estimation systems for clinical mobility assessment. Our iPad-based platform achieves 30fps inference using CoreML and implements CDC STEADI fall-risk protocols, enabling objective measurement of gait, balance, and functional mobility. This work led to a US Patent (12,343,138 B2) for AI-powered mobility assessment.

I also conduct research at TeCSAR Lab at UNC Charlotte, where I’ve contributed to self-supervised human motion foundation models and scalable intelligent video surveillance systems for the AIoT.

My research interests include:

Health Sensing & Biomedical Signal Processing — Time-series analysis from wearable sensors, clinical measurement validation
On-Device ML — CoreML, TensorFlow Lite, edge inference optimization, model compression
3D Human Pose Estimation — Monocular video to skeleton, motion understanding, action recognition
Self-Supervised Learning — Foundation models for human motion, discrete tokenization, Vision Transformers
Edge AI & Embedded Systems — Real-time inference on resource-constrained devices, FPGA acceleration

I earned my Ph.D. in Electrical & Computer Engineering from UNC Charlotte, where I developed lightweight temporal convolutional networks for time-series classification on microcontrollers and contributed to algorithm/architecture co-design for FPGA-based DNN acceleration.

I have published 12+ peer-reviewed papers across IEEE and ACM venues, accumulating 645+ citations (h-index: 11). My work has been featured in IEEE Transactions on Intelligent Transportation Systems, IEEE Internet of Things Journal, and ACM Transactions on Embedded Computing Systems.

news

Jul 01, 2025	US Patent Granted: AI Powered Mobility Assessment System (US 12,343,138 B2) — Enabling automated fall risk evaluation and real-time exercise coaching via 3D skeleton modeling from monocular video.

selected publications

IEEE TITS

DeepTrack: Lightweight Deep Learning for Vehicle Trajectory Prediction in Highways

Vinit Katariya, Mohammadreza Baharani, Nichole Morris, and 2 more authors

IEEE Transactions on Intelligent Transportation Systems, 2022

Abs HTML Code

This paper presents DeepTrack, a lightweight deep learning approach for vehicle trajectory prediction in highways using Temporal Convolutional Networks with depthwise separable convolutions, achieving state-of-the-art accuracy with significantly reduced computational complexity for IoT deployment.
IEEE IoT

REVAMP2T: Real-Time Edge Video Analytics for Multicamera Privacy-Aware Pedestrian Tracking

Christopher Neff, Matias Mendieta, Shrey Mohan, and 3 more authors

IEEE Internet of Things Journal, 2019

Abs HTML Code

An end-to-end scalable intelligent video surveillance system enabling privacy-aware multi-camera pedestrian tracking with real-time edge inference across distributed infrastructure on NVIDIA Jetson Xavier.
IEEE IoT

Ancilia: Scalable Intelligent Video Surveillance for the AIoT

Armin Danesh Pazho, Christopher Neff, Ghazal Alinezhad Noghre, and 4 more authors

IEEE Internet of Things Journal, 2023

Abs HTML

Ancilia presents a scalable intelligent video surveillance architecture for the AIoT, enabling distributed processing across edge and cloud infrastructure with privacy-preserving analytics.
IEEE IoT

Real-Time Deep Learning at the Edge for Scalable Reliability Modeling of Si-MOSFET Power Electronics Converters

Mohammadreza Baharani, Mehrdad Biglarbegian, Babak Parkhideh, and 1 more author

IEEE Internet of Things Journal, 2019

Abs HTML Code

A real-time LSTM-based reliability prediction system for power electronics converters, achieving scalable monitoring on IoT edge devices at 1.87W power budget.
ACM TECS

ATCN: Resource-Efficient Processing of Time Series on Edge

Mohammadreza Baharani and Hamed Tabkhi

ACM Transactions on Embedded Computing Systems, 2022

Abs HTML Code

ATCN presents attention-based temporal convolutional networks enabling fast time-series classification on resource-constrained embedded devices with formalized hyperparameters for application-specific architecture tuning. Deployed on Cortex-M7 consuming only 49% of 320KB RAM and 15% of 1MB flash.
arXiv

MoFM: A Large-Scale Human Motion Foundation Model

Mohammadreza Baharani, Ghazal Alinezhad Noghre, Armin Danesh Pazho, and 2 more authors

arXiv preprint arXiv:2502.05432, 2025

Abs HTML

A self-supervised human motion foundation model using discrete motion tokenization and Vision Transformer pretraining, achieving competitive action classification (89.4%), one-shot recognition (57.4%), and anomaly detection (76.56% AUC-ROC) with a single unified backbone.