Human activity recognition (HAR) is a process of identifying the daily living activities of an individual using a set of sensors and appropriate learning algorithms. Most of the works on HAR are done using a mix of sensor data that is collected in a simulated environment, and due to that, the real-time recognition suffers. This article proposes an efficient adaptive batch size-based-CNN-LSTM model for recognizing different human activities in an uncontrolled environment. It uses adaptive batch sizes from 128 to 1024 for iterative model training and validation. The proposed model can handle imbalanced classes and un-normalized data efficiently. A state-of-art HAR dataset is also generated in an open environment to get the activity data of an uncontrolled scenario. With minimal data preprocessing and data augmentation, the model is tested, and the proposed model managed to get the highest accuracy of 99.29% with an average loss of 0.08 $\pm$ 0.136%. The presented model is also tested with two public datasets named- mHealth and MotionSense and achieved an accuracy of 99.5% and 99.8%, respectively. The proposed model outperforms all the previous benchmarks and approaches by a good margin.

Traditional methods of Human Activity Recognition (HAR) do not take into consideration physical attributes of human subjects such as height, weight, gender etc. Thus, a particular recognition model does not perform consistently for different subjects having diverse physical attributes. In this paper, we propose a novel Physique-based HAR method that takes care of this problem and provides better accuracy in comparatively less time. Raw sensor data are acquired from inbuilt accelerometer and gyroscope sensor modules of smartphones. After pre-processing of the collected data physique-based datasets were prepared, based on the similarities of subjects’ physique. Analysis is done on both of these physique-based datasets and normal traditional dataset with the help of various machine learning algorithms. The work not only identifies the suitable learning algorithms for HAR but also shows that the proposed physique-based method outperforms the traditional HAR approach for both publicly available dataset and our own generated dataset, with an accuracy of 99.88%. Individual activity-wise accuracy results are also compared with several recent benchmarks to show the efficiency of the proposed Physique-based HAR method.

Human Activity Recognition is the process of identifying daily living activities of a person using sensor attributes and intelligent learning algorithms. Identifying complex human activities is tedious, as capturing long-term dependencies and extracting efficient features from the raw sensor data is challenging. This paper proposes an efficient and lightweight hybrid deep learning model for recognizing complex human activities using physiological EMG sensors and enhanced feature learning. The proposed CNN-LSTM incorporates multiple 1-dimensional convolution layers for spatial feature extraction and then feeds the generated feature maps to the recurrent layers to identify long-term temporal dependencies. Incorporating physiological sensor-based raw EMG dataset and minimal pre-processing, we trained and tested our proposed model and achieved the highest accuracy of 84.12% and an average accuracy of 83%. The proposed model outperformed the benchmark models with optimal performance margins and generalized the patterns in significantly less computational time than other deep learning models.

Human Activity Recognition (HAR) is the process of identifying daily living activities using a set of sensors and optimal learning algorithms. It is a convoluted process, as there is no straightforward way to associate human action with the induced sensor data. Most of the work on HAR is done on highly augmented and pre-processed data. It optimizes performance but introduces intense data pre-processing and feature engineering overhead in real-time activity recognition. This paper proposes an efficient and lightweight CNN-LSTM model for enhanced activity classification on raw sensor data in an uncontrolled environment. It used convolution cum memory functionalities of the CNN-LSTM layer to extract spatial and temporal features for distinguished feature analysis and also classifies different human activities using dense classification layers. A state-of-the-art inbuilt smartphone sensor-based HAR dataset is also generated for six daily living activities in an uncontrolled environment for getting real-world activity data. With zero data pre-processing and augmentation, our proposed Conv1D-based CNN-LSTM model out-completed all the incorporated conventional models and achieved the highest accuracy of 98% in optimized computational time. Also, the loss on real-world test data was minimum, with the loss perimeter advantage of 68% and 158% with LSTM and ANN models, respectively.

Human Activity Recognition (HAR) is one of the most exploited areas in the field of artificial intelligence and wearable sensor technology. Most of the work on HAR classifies simple to complex human activities but does not manage to recognize the actions of an individual with distinct physical abnormalities. This paper proposes an efficient and lightweight hybrid deep learning model (ConvLSTM) to classify human activities with knee abnormalities. A state-of-the-art publicly available physiological sensor-based HAR dataset with knee abnormalities is incorporated to validate the proposed framework. A minimal data pre-processing pipeline was also introduced to prepare the dataset for categorization without interfering with its raw characteristics. Further, optimal hyperparameters with an efficient time-step-based data segmentation method were developed for automatic feature engineering in spatial and temporal frequency domains. Moreover, incorporating regularization and low-cost mathematical functions resulted in a sparse and prominent feature space, improving performance by overcoming overfitting and optimizing computational times. The proposed model achieved the highest accuracy of 99.21% and an average accuracy of 99% in comparatively optimized computational time than the benchmark and state-of-the-art models.