Research On Human Activity Recognition Based On Smartphone Sensor Data

Human activity recognition is a hotspot in the field of Pattern Recognition,these kinds of researches are based on video data and smartphone sensor data.With the aggravation of aging society,the health of the elderly are more and more concerned,using the human activity recognition technology for the elderly to get the high degree of intelligent medical care can provide accurate assessment information for the caregiver.This paper mainly focuses on the problems of large amount of experimental data and poor classification results in human activity recognition,based on the real data,we propose two kinds of innovative algorithms:human activity recognition based on multi-stage continuous hidden Markov model and human activity recognition based on sparse locality preserving projection combined with random forest ensemble classifier.The new algorithms effectively solve the difficulties encountered in the current activity recognition research.The main innovations of this paper are as follows:(1)The traditional continuous hidden Markov model faces the problem of low accuracy in the process of activity recognition.Based on the hierarchical characteristics of human activities and timing,diversity,continuity of the sensor data,we propose a new algorithm of human activity recognition based on three-stage continuous hidden Markov model.The experimental results show that the proposed algorithm not only can clearly identify the misclassified categories,but also solve the problem of low recognition rate,especially the accuracy of classification of confuse activities.(2)Sparse locality preserving projections(SpLPP)are applied in human activity recognition of continuous hidden Markov model for the first time.SpLPP is exploited to determine the optimal feature subsets,it can extract more discriminative activities features from the sensor data compared with locality preserving projections.The experimental results show that the new algorithm is effective.(3)Some studies have used random forest(RF)classifiers for human activity recognition on smartphone sensor data since the recognition rate is ensemble classifier better than the single classifier.However their approaches did not make full use of the cutting-edge technology.So,we propose use the SpLPP as the method of reduce dimension.This method effectively solves the problem of the large number of features and reduced the time complexity of the experiment.The results of the new algorithm also show that the overall recognition rate of activity recognition is improved significantly by using SpLPP.