A survey on human activity recognition using CNN and LSTM

Abstract

Human Activity Recognition (HAR) plays a massive role inside the everyday life of people due to its capability to investigate tremendous high-degree data approximately human interest from wearable or stationary devices. An extensive quantity of studies has been completed on HAR and numerous techniques primarily based on deep gaining knowledge of and device analyzing were exploited through the research community to categorize human sports. The principle motive of this evaluation is to summarize contemporary works based on a big range of deep neural networks structure, mainly convolution neural networks (CNNs) for human hobby reputation. The reviewed systems are clustered into 4 classes relying on the use of input devices like multimodal sensing gadgets, smart phones, radar, and vision devices. This assessment describes the performances, strengths, weaknesses, and the used hyper parameters of CNN architectures for every reviewed device with an evaluate of available public statistics property in addition, a communicate with the cutting edge demanding situations to CNN based HAR systems is furnished. sooner or later, this assessment is concluded with a few functionality future guidelines that might be of outstanding assistance for the researchers who would really like to make contributions to this subject.