Abstract:Video-based gait analysis can be defined as the task of diagnosing pathologies, such as ataxia, using videos of patients walking in front of a camera. This paper presents a graph convolution network called AtGCN for detecting ataxic gait and identifying its severity using 2D videos. The problem is especially challenging as the deviation of an ataxic gait from a healthy gait is very subtle. The datasets for ataxic gait detection are also quite small, with the largest dataset having only 149 videos. The paper addresses the first problem using special spatiotemporal graph convolution that successfully captures important gait-related features. To handle the small dataset size, a deep spatiotemporal graph convolution network pre-trained on an action recognition dataset is systematically truncated and then fine-tuned on the ataxia dataset to obtain the AtGCN model. The paper also presents an augmentation strategy that segments a video sequence into multiple gait cycles. The proposed AtGCN model then operates on a graph of body part locations belonging to a single gait cycle. The evaluation results support the strength of the proposed AtGCN model, as it outperforms the state-of-the-art in detection and severity prediction with an accuracy of 93.46% and a MAE of 0.4169, respectively.

What problem does this paper attempt to address?

### Problems Addressed by the Paper This paper aims to address the problem of detecting ataxic gait and its severity through video analysis. Specifically: 1. **Ataxic Gait Detection**: Ataxia is a disease that affects the body's coordination, manifesting as abnormal gait. Early detection of ataxia is crucial for disease management and treatment. However, current assessment methods mainly rely on visual inspection by professionals, which is not only time-consuming but also prone to errors. 2. **Small Dataset Size**: The datasets used for ataxic gait detection are very small, with the largest dataset containing only 149 video records. This makes it difficult for traditional deep learning models to train and generalize. 3. **Subtle Gait Features**: The differences between ataxic gait and normal gait are very subtle, making automatic detection highly challenging. ### Solution To address the above issues, the paper proposes a model based on Graph Convolutional Network (GCN), called AtGCN. The main features of this model include: 1. **Special Spatio-Temporal Graph Convolution**: AtGCN uses special spatio-temporal graph convolution to capture important gait features. These features include the positions and movement patterns of various body parts. 2. **Data Augmentation Strategy**: To tackle the problem of small dataset size, the paper proposes a data augmentation strategy by segmenting video sequences into multiple gait cycles. Each gait cycle is converted into a graph structure, which serves as the input to the model. 3. **Pre-training and Fine-tuning**: The AtGCN model is obtained by systematically truncating and fine-tuning a deep spatio-temporal graph convolutional network pre-trained on an action recognition dataset. This leverages the knowledge of the pre-trained model to improve performance on the small dataset. ### Experimental Results Experimental results show that the AtGCN model performs excellently in ataxic gait detection and severity prediction. Specifically: - **Detection Accuracy**: The detection accuracy of the AtGCN model reaches 93.46%, significantly higher than existing methods. - **Severity Prediction**: The mean absolute error (MAE) of the AtGCN model is 0.4169, also better than existing methods. ### Conclusion By proposing the AtGCN model, this paper successfully addresses the key issues in ataxic gait detection and severity prediction. The model not only outperforms existing methods in terms of performance but also provides new directions for future research.

AtGCN: A Graph Convolutional Network For Ataxic Gait Detection

G-GCSN: Global Graph Convolution Shrinkage Network for Emotion Perception from Gait

Condition-Adaptive Graph Convolution Learning for Skeleton-Based Gait Recognition

Skeleton-based abnormal gait recognition with spatio-temporal attention enhanced gait-structural graph convolutional networks

Gait Recognition Using Multichannel Convolution Neural Networks

FP-GCN: Frequency Pyramid Graph Convolutional Network for Enhancing Pathological Gait Classification

Classification of Abnormal Gait Based on Improved Two-Stream CNN

Cross-Spatiotemporal Graph Convolution Networks for Skeleton-Based Parkinsonian Gait MDS-UPDRS Score Estimation

Stgcn-pad: a spatial-temporal graph convolutional network for detecting abnormal pedestrian motion patterns at grade crossings

Spatiotemporal smoothing aggregation enhanced multi-scale residual deep graph convolutional networks for skeleton-based gait recognition

GLGait: A Global-Local Temporal Receptive Field Network for Gait Recognition in the Wild

GaitCTCG: cross-view gait recognition via cascaded residual temporal shift and comprehensive multi-granularity learning

Gait Recognition Based on 3D Skeleton Data and Graph Convolutional Network

GAS-GCN: Gated Action-Specific Graph Convolutional Networks for Skeleton-Based Action Recognition.

Fall Detection Method for Infrared Videos Based on Spatial-Temporal Graph Convolutional Network

GaitRec-Net: A Deep Neural Network for Gait Disorder Detection Using Ground Reaction Force

Auto-Correlation and Channel Attention Enhanced Deep Graph Convolution Networks for Gait Phase Prediction Based on Multi-IMU System

Three-dimensional convolutional neural networks applied to video sensor-based gait recognition

Multimodal Gait Abnormality Recognition Using a Convolutional Neural Network–Bidirectional Long Short-Term Memory (CNN-BiLSTM) Network Based on Multi-Sensor Data Fusion

Graph Instinctive Attention Convolutional Network for Skeleton-Based Action Recognition.

GaitMGL: Multi-Scale Temporal Dimension and Global–Local Feature Fusion for Gait Recognition