An Efficient Ensemble Framework for Human Gait Recognition Using CNN-LSTM With Extra Tree Classifier and Smartphone Sensors in Real-World Environment

Abstract

Gait recognition is a biometric technology that identifies individuals based on their unique way of walking. Most of the work on human gait recognition (HGR) systems has minimal user records and is performed in a closed simulated environment, which hampers the performance in a real-world scenario. This letter presents an efficient ensemble framework using a hybrid deep learning network (convolutional neural network-long short-term memory) with an extra tree classifier (ETC) for HGR in a real-world environment. The proposed model effectively extracts low-level spatial and temporal features from the sensor data for meaningful pattern generation and classifies them using multiple decision trees present in the ensemble ETC. A State-of-the-Art HGR dataset has also been developed for a diverse set of users in uncontrolled environments in real-world environments using built-in smartphone sensors. The proposed model achieved an average performance accuracy of 99.10% and optimal precision, recall, and F1-score, outperforming all the benchmark models with optimal performance margins in lower computational times.