Acknowledgment of patient in sense behaviors using bidirectionalConvLSTM

Abstract

Recognizing patient activity in real‐time from video or images collected by a CCTV camera available in the hospital during a Covid‐19 situation has proven challenging. The dilemma of patient activity recognition is identifying and recognizing a patient's various actions in a series of videos. The process presented in our paper needs to achieve unrestricted, generic behavior in videos. Detecting events in any video is often difficult because we use Bidirectional ConvLSTM to create a robust patient in the sense behaviors (PSB) framework capable of eliminating certain barriers. To begin this paper by proposing a new Bidirectional ConvLSTM for establishing a stable PSB scheme. Our proposed model is capable of accurately predicting patient's behaviors like seated, standing, and so on. Using Bidirectional ConvLSTM, learning information from a pre‐trained model is an excellent place to start for rapidly developing a new PSB system using a current PSB database, as both the source and target datasets are critical. All parameters are frozen in a pre‐trained PSB device. Then, using the UCI and HMDB51 dataset to train the model, variables and local relations are progressively fixed. A novel PSB framework is developed using the target dataset. Relevant tests are conducted using commonly used research indices to assess prediction precision accuracy. They acknowledge six patient's behavior with a weighted accuracy rate of 92%. For recognizing novel activity, laying, the precision of a corresponding prediction is the best, 91%, of all six test results. The proposed work uses bidirectional ConvLSTM with modified activation layers to sense the patients' behavior. This article may be a patient activity recognition system to identify an individual. It takes a clip of COVID‐19 patients as input and looks for matches inside the hold‐on images.