Thermopile-Based Occupancy Estimation

Abstract

Code and data for benchmarking rule-based, feature-based ML (logistic regression, SVM, random forest), and temporal-CNN occupancy counters on low-resolution thermopile array sensors (Grid-EYE 8×8 and HTPA 32×32), under clean and thermal-interference conditions. Includes raw per-frame sensor CSVs for 0–4 occupants across both sensors and conditions, the full training/evaluation pipeline, pretrained model weights, and the generated result tables and figures reported in the accompanying IEEE Access paper. Provided to support reproducibility of the paper's in-domain accuracy and cross-domain robustness benchmarks.