Gait Recognition using Double-Window and CNN Classification on Freestyle Walks
- Resource Type
- Conference
- Authors
- Limcharoen, Piya; Khamsemanan, Nirattaya; Nattee, Cholwich
- Source
- 2018 Joint 10th International Conference on Soft Computing and Intelligent Systems (SCIS) and 19th International Symposium on Advanced Intelligent Systems (ISIS) Soft Computing and Intelligent Systems (SCIS) and 19th International Symposium on Advanced Intelligent Systems (ISIS), 2018 Joint 10th International Conference on. :1231-1237 Dec, 2018
- Subject
- Communication, Networking and Broadcast Technologies
Computing and Processing
Robotics and Control Systems
Signal Processing and Analysis
Legged locomotion
Feature extraction
Cameras
Gait recognition
Skeleton
Shape
Streaming media
Human identity
Gait Recognition
Human identification
Kinect
Biometrics
Freestyle walks
View independent
Sliding window
Convolution Neural Network (CNN)
k-Nearest Neighbor
- Language
In this paper, we introduce a new technique for human identification using the gait of freestyle walks. Gait represents locomotions in animals and human beings. For a human, it is called a bipedal walk or a bipedal run. Each person has his or her own unique character of gait movements. We propose a gait recognition technique based on the concept of double-windows. A double-window is a series of consecutive usable frames in one walking sequence. One double-window is divided into two equal groups of frames. Gait features are extracted from double-windows using sliding window techniques. The data size of gait features of a walking sequence based on the proposed technique is smaller than those extracted using existing methods. The proposed double-window feature extraction technique also solves a viewpoint issue. Convolutional Neural Networks and k-Nearest Neighbor are used in the classification process. The accuracy of our proposed double-window technique is 99.33% which outperforms other techniques.