HotFoot: Foot-Based User Identification using Thermal Imaging

Title of the Paper

HotFoot: Foot-Based User Identification Using Thermal Imaging

Paper Information

• Domain: Foot-based user identification using thermal imaging
• Keywords: Footprint, user identification, thermal imaging, biometric identification, smart environments, machine learning, gradient boosting, classifier, thermal features, visual features

Research Background and Problem

• Problem or Challenge:

  • Current smart environments primarily rely on user behavior rather than user identity, limiting the development of personalized applications.
  • Many existing user identification technologies (e.g., fingerprint recognition, facial recognition) require user interaction or consume significant sensor resources, with restricted applicability in certain scenarios.
  • While foot-based identification has potential, many approaches require users to remove shoes or socks, limiting practical application.

• Significance:

  • Improving user identification technology enables smart environments to automatically adapt to individual preferences, enhancing security and user experience.
  • Foot-based identification has the potential to become a non-intrusive, continuous identification method, and thermal imaging offers new possibilities.

• Research Motivation and Related Work:

  1. Foot interaction technologies have been extensively studied, but research on foot-based identification remains limited.
  2. Biometric identification technologies have expanded to behavior-based multimodal methods, such as gait analysis and the use of thermal imaging devices.
  3. Thermal imaging has been effectively applied in user gesture detection and facial recognition, but its application in user identity recognition remains scarce.

Solution

• Main Method or Solution:

  • A user identification method combining thermal imaging and visual features is proposed, exploring foot thermal radiation and residual heat traces.
  • Thermal imaging cameras are used to capture foot thermal radiation data and visual images, inferring user identity based on foot heat distribution and geometric characteristics.

• Innovations:

  1. A novel foot biometric identification method combining thermal imaging and visual features is proposed.
  2. A prototype system was implemented, and a publicly available dataset was collected to facilitate further research.
  3. The method's accuracy was validated in various scenarios (different floor materials and footwear conditions), achieving a maximum AUC score of 98.9%.

• Implementation Steps:

  1. Data Collection: In a laboratory environment, data from 21 participants were collected under three floor materials (carpet, wooden flooring, linoleum) and three footwear conditions (socks, personal shoes, standard shoes).
  2. Feature Extraction: Thermal features (temperature distribution, heat dissipation time, etc.) and visual features (geometric shape, etc.) of the foot were extracted using computer vision techniques.
  3. Classification Modeling: A random forest classifier was constructed to predict user identity, divided into four classification scenarios based on footwear and floor conditions.
  4. Performance Evaluation: Metrics such as AUC and F1 scores were used to evaluate the performance of different models.

Research Results

• Specific Results:

  1. A total of 567 thermal imaging data samples, including visual and thermal characteristics, were collected to form a publicly available dataset.
  2. The study demonstrated that classifiers combining visual and thermal features achieved the best performance in user identification (maximum AUC of 98.9%), significantly outperforming single-feature approaches.
  3. Thermal features outperformed visual features (average AUC of 0.91+ vs. 0.72+), showing robustness against changes in user posture or environmental lighting.

• Advantages:

  • The method requires no additional user interaction, enabling lightweight and continuous identification.
  • Compared to traditional biometric methods, it offers greater flexibility in diverse environments (e.g., combinations of different floors and footwear).

• Experimental or Evaluation Results:

  1. Under fully constrained conditions (same floor and footwear), the classifier performed best.
  2. Under fully unconstrained conditions (different footwear and floors), the classifier's performance slightly decreased but remained effective (AUC of 91.7%).
  3. Floor material had minimal impact on identification performance, while footwear type (e.g., socks vs. shoes) had a more significant effect.

• Limitations and Future Directions:

  1. Current experiments were conducted in controlled environments, lacking coverage of large-scale real-world scenarios.
  2. The sample size was limited (21 participants), requiring validation on larger datasets for scalability.
  3. Testing did not include disabled users (e.g., individuals with single limb or chronic cold foot conditions).
  4. Future research could explore more complex thermal pattern recognition methods to enhance robustness and address spoofing attacks in security verification applications.