View Proposal

---

Proposer

Inkar Zhumay

Title

Device-Free Human Activity Recognition from WiFi Channel State Information

Goal

To determine how well WiFi channel state information supports device-free activity recognition when the system is evaluated across different people, rooms and device placements rather than within a single controlled setting.

Description

For Aktobe Students Only. Build a recognition pipeline over WiFi CSI measurements: denoising and phase sanitisation, subcarrier selection, feature extraction or learned representation, and classification of a small set of coarse activities. The central experiment is generalisation, not accuracy within a fixed setup — evaluate under leave-one-subject-out, leave-one-room-out and leave-one-orientation-out protocols, and compare against the within-setting result to quantify the drop. Reported accuracies in this field are typically within-environment and collapse on relocation; measuring that collapse honestly is the contribution. Evaluate whether domain adaptation, adversarial feature alignment or simple normalisation recovers any of the lost performance. If hardware permits, collect a small dataset with commodity network cards to validate the findings outside public benchmarks.

Resources

Background

Url

Difficulty Level

Moderate

Ethical Approval

None

Number Of Students

1

Supervisor

Inkar Zhumay

Keywords

wifi sensing, human activity recognition, wireless sensing, deep learning

Degrees

Bachelor of Science in Computational Sciences and Software Engineering