Multiple Task Human Gait Analysis and Identification: Ensemble Learning Approach

Abstract

Gait analysis refers to the study of human locomotion which plays an important role in clinical assessment for the identification of gait abnormality for neurological disorder persons and for athletes. This can also be used in biometrics as it is unique and difficult to hide. The human gait is considered a very unique for each subject. This chapter tried to answer the question, what are the significant features to identify the different activity of human? The second research question which this chapter has addressed was how one can efficiently identify the different activity of walking and provide the generic solution. In this chapter, artificial neural network is used to classify the human gait and compared with ELM. The dataset contains data collected from the various tasks like: walking at natural speed (N), walking very slow (XS), walking slow (S), walking medium (M), walking fast (L), walking on toes (T), walking on heels (H), stair ascending (U), and stair descending (D). These nine behaviors are classified on the basis of the following features like Pelvic Ant/Posterior Tilt, Hip Flex/Extension, Hip Ad/Abduction, Hip Internal/External Rotation, Knee Flexion/Extension, and Ankle Dorsi/Plantar flexion. The algorithm used for the classification is ELM as it provides good classification results in less computational time. The performance is also compared with SVM and KNN algorithms. This chapter also incorporates PCA technique to determine the top gait features. The results showed that the classification accuracy of ELM is better than SVM and KNN. To provide the generic solution and less complex model the ensemble learning is being explored. The combination of different classifier provides the average performance which avoid over fitting and less dependence on hyper-parameter. The ensemble learning technique has provided the much need generic to our proposed solution for multi activity gait classification.