CHAPTER 1INTRODUCTION1.1 Human Activity Recognition Human Activity Recognition is the process of classifying sequences of accelerometer data recorded by smartphones into known movements. It is the problem of what actually a person is doing with the help of sensors. The Movements are indoor activities such as standing, walking, lying, running, jogging, sitting. Sensors are embedded in smartphones to record the movement of humans. There are a lot of sensors embedded in smartphones. The sensors used for human activity recognition are accelerometer and gyroscope.
Widely used sensor for the HAR is accelerometer sensor. Accelerometer sensor can measure acceleration in one, two or three axes. The Sensors record the accelerometer data in three dimensions called tri-accelerometer. HAR is active research areas in human-computer interaction. The idea is that when the subject activity is known and recognized, an intelligent system can offer assistance. Most of the daily tasks of humans can be automated if they are recognized by HAR system. HAR system can be supervised or unsupervised.
A Supervised HAR system requires some former training with consecrate datasets. An unSupervised HAR system is configured with rules while being developed.HAR is considered as an important element in various scientific fields such as surveillance, health care, etc. Activity Recognition based on sensors incorporate with the evolving sensor network area with machine learning techniques to model human activities. Mobile devices provide sensor data and calculation power to enable physical activity recognition and to estimate the energy consumption of human in day-to-day life. The demand for recognizing human activities have grown in the health domain. Many studies found that wearable sensors have very low error rate for predicting the activity. This project uses sensors that are embedded in mobile devices to recognize human activity as many smartphones come with in-built sensors. Smartphones have become more important in humans life. The activities performed by a user can be detected from the values of an accelerometer. The accelerometer values for each activity show a different pattern. We are developing this application for the purpose of health care where we are calculating the calories burnt based on the activity performed by a user is recognized.1.2 Deep Learning Deep Learning is a branch of machine learning which is based on a set of algorithms. In the simple case, when the input layer receives an input it passes on a modified version of the input to the next layer. In a neural network, there are many layers between the input and the output layer, and the algorithm uses multiple processing layers, composed of multiple linear and non-linear transformations. The proposed system uses deep learning techniques for human activity recognition. Fig 1.1 Working of Neural Network Deep neural networks are much harder to train than light neural networks. The type of deep network used in this project is deep convolutional network and recurrent neural network. A Con-volutional Neural Network (CNN) is comprised of one or more convolutional layers and then followed by one or more fully connected layers. Convolutional neural net- works (CNN, or ConvNet) were inspired by biological processes and are variations of multilayer perceptron designed to use minimal amounts of pre-processing. A CNN consists of a number of convolutional, subsampling layers and optionally followed by fully connected layers. The input to a convolutional layer is a m x m x r image where m x m is the height and width of the image and r is the number of channels. The convolutional layer will have k filters of size n x n x q where n is smaller than the dimension of the image and q can either be the same as the number of channels r or smaller and may vary for each kernel. Convolutional neural networks use three ideas: local receptive fields, shared weights, and pooling. Every neuron in the first hidden layer will be connected to a small region of the input neurons. That small region in the input image is called the local receptive field for the hidden neuron. It’s a little window on the input pixels. We slue the local receptive field across the entire input image. For each local receptive field, there is a unique hidden neuron in the first hidden layer. The map from the input to hidden layer is called a feature map. The weights that define the feature map are called the shared weights. And the bias defining the feature map in this way are called the shared bias. The shared weights and bias are often said to define a kernel or a filter. A pooling layer takes each feature map output from the convolutional layer and prepares a distil feature map. It simplifies the information in the output from the convolutional layer. A common procedure for pooling is known as max-pooling. LSTM is an artificial RNN used in deep learning field. It also has feedback connections. It not only process the single data points but also process the entire sequence of data. A LSTM unit is composed of cell, an input , output and forgot gate. Since there is can be lab between the events in the time series LSTM is mostly suited for classifying, processing and making predictions based on the time series data.145415453521 Fig 1.2 LSTM building block RNN is the only one with the internal memory. Because of the internal memory they are able to remember important content about the input what they have received which enables them in predicting what’s coming next. It produces predictive results in sequential data that other algorithms can not produce.1.3 Organization of the Project Report The project report is organized as follows:In Chapter (2), we discuss about the problem statement and our solution to the problem. The same chapter also deals with the other existing state of arts. The Chapter that follows i.e chapter (3) consists of the details on the literature survey of the papers to the problem statement and the proposed solution. In Chapter (4), we present the System Overview and Proposed system in the form of Data flow diagram and the sequence diagram. The next chapter, chapter (5)gives the requirements and detains about the implementation of the proposed system. Chapter 6 deals with the testing of the system and their results. In Chapter (7), we discuss about the influencing parameters and their effect on the system. The same chapter deals with establishing the optimal parameters for the system. The chapter (8) concludes the paper along with mention of the Future Enhancements.Chapter (9) is details about the references made during the development of the system. The other supporting information and the source code are gathered in the Appendix. CHAPTER 2Problem Statement And Proposed Solution2.1 Problem Statement