KinectGaitNet: Kinect-Based Gait Recognition Using Deep Convolutional Neural Network.

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Gait recognition is valuable for surveillance, security, and healthcare.
• Existing methods often require handcrafted features or complex processing.
• Identifying individuals by gait offers unobtrusive and cost-effective biometric solutions.

Purpose of the Study:

• To propose KinectGaitNet, a novel convolutional neural network for Kinect-based gait recognition.
• To develop a deep learning model that directly utilizes 3D joint coordinate data.
• To achieve state-of-the-art performance in gait recognition without manual feature engineering.

Main Methods:

• A unique 3D input representation of body joint coordinates over the gait cycle was created.
• The KinectGaitNet model, a convolutional neural network, was trained directly on this 3D representation.
• Residual learning was employed to enhance accuracy and prevent performance degradation.

Main Results:

• KinectGaitNet achieved 96.91% accuracy on the UPCV dataset and 99.33% on the KGB dataset.
• The model surpassed all previous state-of-the-art methods for Kinect-based gait recognition.
• The proposed method demonstrated superior performance with fewer parameters and reduced inference time.

Conclusions:

• KinectGaitNet represents a significant advancement in deep learning-based gait recognition.
• The novel 3D input representation and direct training approach are highly effective.
• This method offers a more efficient and accurate solution for real-world gait recognition applications.