Researchers claim mmWave biometrics enable privacy preserving emotion recognition dataset

Researchers have released what they describe as the first publicly accessible emotion recognition dataset based on millimetre‑wave (mmWave) radar.

The advantages are that this offers a contact‑free and privacy‑preserving alternative to traditional biometric sensors, the study’s authors note. Biometric Update has an explainer on emotion recognition and its relation to biometrics.

The dataset captures emotional responses from fifteen participants using three simultaneous signals: mmWave radar readings, photoplethysmography (PPG) pulse data and galvanic skin response (GSR).

Participants also provided subjective ratings using the Self‑Assessment Manikin (SAM), a standard tool for measuring emotional valence, arousal and dominance.

Unlike conventional emotion recognition datasets, which rely on wearable devices or video and audio recordings, the mmWave approach collects physiological information without touching the body or capturing identifiable imagery.

The radar tracks subtle chest‑wall movements caused by breathing and heartbeat, allowing researchers to extract vital‑sign patterns without raising the privacy concerns associated with cameras or microphones.

The team used validated film clips to induce emotional states and then analyzed the resulting signals to confirm both the effectiveness of the stimuli and the quality of the biometric data. They say the dataset opens new avenues for research, including cross‑subject emotion recognition, multimodal fusion techniques and comparisons between radar‑based and traditional physiological signals.

It has to be said here that emotion recognition systems are considered highly dubious, if not outright unscientific, due to the difficulty of objectively assessing emotions. Emotion recognition technology has also been criticized for violating fundamental rights. This is especially the case in the European Union where the EU AI Act prohibits the development, deployment and placement of emotion recognition systems in the EU market intended for workplaces and educational institutions.

However, the academic study here is highlighting the collection of emotion recognition datasets, using new techniques, rather than developing emotion recognition. The study, “An emotion recognition dataset using millimeter wave radar and physiological reference signals,” is found in the academic journal Nature and was conducted on college students.

But the paper concedes limitations, such as the small sample size of 15, healthy college-age students, who are based in China. “Future studies should aim for larger, more demographically diverse populations to ensure better generalization,” the paper says.

More generally, it is difficult to objectively know whether watching film clips really does induce emotional states, as the researchers claim. For example, the researchers used clips from Hollywood movies including The Shawshank Redemption and Wall-E to induce sadness and happiness, respectively. The paper does not address this, probably because it’s a larger issue beyond its purview.

Because mmWave sensing is unaffected by lighting conditions, ambient noise or user discomfort, the researchers argue it could form the basis of a new class of biometric systems that infer emotional states while minimizing privacy risks.

The dataset is intended for studies ranging from vital‑sign extraction to investigations of individual differences in emotional responses, and the authors note that no comparable mmWave‑based emotion‑recognition dataset has previously been available to the research community. The study can be read in full here.

Article Topics

data privacy  |  emotion recognition  |  research