Deep CSI Learning for Gait Biometric Sensing and Recognition (1902.02300v1)

Abstract: Gait is a person's natural walking style and a complex biological process that is unique to each person. Recently, the channel state information (CSI) of WiFi devices have been exploited to capture human gait biometrics for user identification. However, the performance of existing CSI-based gait identification systems is far from satisfactory. They can only achieve limited identification accuracy (maximum $93\%$) only for a very small group of people (i.e., between 2 to 10). To address such challenge, an end-to-end deep CSI learning system is developed, which exploits deep neural networks to automatically learn the salient gait features in CSI data that are discriminative enough to distinguish different people Firstly, the raw CSI data are sanitized through window-based denoising, mean centering and normalization. The sanitized data is then passed to a residual deep convolutional neural network (DCNN), which automatically extracts the hierarchical features of gait-signatures embedded in the CSI data. Finally, a softmax classifier utilizes the extracted features to make the final prediction about the identity of the user. In a typical indoor environment, a top-1 accuracy of $97.12 \pm 1.13\%$ is achieved for a dataset of 30 people.

Deep CSI Learning for Gait Biometric Sensing and Recognition

The paper "Deep CSI Learning for Gait Biometric Sensing and Recognition" presents an advanced approach to improving gait recognition accuracy using Channel State Information (CSI) from WiFi signals. The authors address the limitations of existing systems and propose a deep learning framework that significantly enhances user identification accuracy. This summary provides an overview of the methodology, results, and potential implications of this research.

Overview of the Methodology

The authors identify significant limitations in traditional CSI-based gait recognition systems, which often rely on hand-crafted features and basic machine learning models. These systems have historically achieved limited accuracy and scalability, being effective only for small groups of subjects. To overcome these challenges, the authors propose a comprehensive pipeline that leverages deep learning to automatically learn relevant features from CSI data.

1. Data Preprocessing: The raw CSI data undergoes several preprocessing steps. Mean imputation is used to handle missing data resulting from dynamic modulation and coding schemes in WiFi transmissions. A Hanning window function denoises the data by filtering out high-frequency noise while retaining the core gait features. Additionally, mean centering and normalization are performed to enhance the data quality for subsequent deep learning processes.
2. Deep Learning Framework: The centerpiece of the proposed system is a residual deep convolutional neural network (DCNN). This model architecture is aimed at extracting hierarchical features from the sanitized CSI data. By opting for a deep learning model over traditional feature engineering methods, the authors eliminate the need for manually crafted feature sets and instead allow the network to identify and learn discriminative patterns in the data.
3. Data Collection and Model Training: The authors collected CSI data from 30 users in an indoor setting using a standard WiFi setup consisting of off-the-shelf equipment. Their dataset supports training and evaluation of the model, achieving an impressive top-1 accuracy of 97.12% across 30 subjects. This performance gain is attributed largely to the effective preprocessing techniques and the model's ability to automatically decipher intricate features associated with gaits.

Results and Implications

The results presented demonstrate a substantial improvement over previous methods, with the proposed system achieving a top-1 accuracy gain of approximately 4% compared to earlier models. This advancement is particularly notable given the scalability to larger datasets and more diverse subjects. The paper also highlights the robustness of the system across different individuals, as reflected in the confusion matrix and classification report findings.

The implications of this work are manifold:

• Practical Applications: The proposed system could be implemented in various real-world security and surveillance applications, where unobtrusive and non-intrusive biometric identification is valuable. Sectors like healthcare may benefit from passive and continuous monitoring of patients through WiFi infrastructures.
• Theoretical Contributions: On a theoretical level, the success of this approach showcases the efficacy of deep learning in tasks traditionally dominated by manual feature engineering. It encourages further exploration into wireless-based biometric recognition using deep models.
• Future Directions: Future research may focus on extending this framework to work in more heterogeneous environments or improving the model's performance under different walking conditions and interference levels. Additionally, transfer learning could be explored to adapt the model to new environments more efficiently.

Conclusion

This paper makes a significant contribution to the field of wireless-based biometric recognition by integrating deep learning techniques with CSI to enhance gait recognition accuracy. The novel application of a residual DCNN to automatically learn gait features from raw CSI data sets a new standard for user identification accuracy in this domain. As research advances, this work will likely inspire additional studies and practical applications leveraging WiFi-based sensing for biometric recognition.