Mobile phones are becoming increasingly advanced and the latest ones are equipped with many diverse and powerful sensors. These sensors can be used to study different position and orientation of the phone which can help smartphone manufacture to track about their customers handling from the recorded log. The inbuilt sensors such as the accelerometer and gyroscope present in our phones are used to obtain data for acceleration and orientation of the phone in the three axes for different phone vulnerable position. From the data obtained appropriate features are extracted using various feature extraction techniques. The extracted features are then given to classifier such as neural network to classify them and decide whether the phone is in a vulnerable position to fall or it is in a safe position .In this paper we mainly concentrated on various case of handling the smartphone and classified by training the neural network.
Human activity recognition system using devices like cameras or microphones have become an active field of research. It has the potential to be applied in different applications such as ambient assisted living. So human activity recognition system has become a part of our daily lives. Smartphones incorporates many diverse and powerful sensors, which can be used for human activity recognition. These sensors include GPS sensors, audio sensors (microphone), vision sensors (cameras), temperature sensors, acceleration sensors (accelerometers), light sensors and direction sensors (magnetic compasses). The data from these sensors can be transferred through wireless communication such as Wi-Fi, 4G and Bluetooth [1]. We have seen that accelerometers and gyroscopes have the most applications as they are the most accurate ones [2]. Studies have shown that activity recognition system using mobile phones are the most extensively used topic in the research domain. Motion-based and location-based activity recognition using in-built sensor and wireless transceivers are the dominating type of activity recognition on mobile phones [3]. Under motion-based activity recognition systems, 3-axis accelerometers are the mostly used sensors available on phones. Most of the studies focus on detecting the locomotion activities, such as standing, walking [4]. Study on various phone positions and orientations and how these positions change different parameters of in-built gyroscope and accelerometer has been limited [5] [6].When a phone is kept at a particular orientation or position there many parameters associated to it. The location of the phone decides a lot about its future [7]. Good results have been obtained in [8] that uses Ameva discretization algorithm and a new Ameva-based classification system to classify physical activity recognition on Smartphones. On comparing the accuracy of human activity recognition, it was found that using only a basic accelerometer gave an accuracy of 77.34%. However, this ratio increased to 85% when basic features are combined with angular features calculated from the orientation of the phone [9]. Human activity recognition using accelerometer was done for some common positions and accuracy was around 91% [10]. Hence an accelerometer alone cannot give very accurate results. But activity recognition significantly increases the efficiency. In contradiction, [11] suggests that this technique still needs a lot of research before it can be used for the general masses. Using few pre-processing techniques, efficiency can be increased too but with many limitations [12]. In this paper, we have focused on different phone positions which are considered to be risky and harmful. Based on these risky positions various parameters such as the roll, pitch and azimuth changes. Whether the phone is at a slipping point or kept on a table or kept on a book. All these factors would decide if the phone will be safe after a certain movement or jerk is applied. And if the jerk moves the phone by a certain distance, will the phone be still safe or there would be a wide change in orientation of the phone which may result in the fall of it. To know all these things beforehand, we have come up with an idea which will tell the user by the change in orientation of the phone that whether the phone is safe or it is in a risky position. The most basic sensors that can be used for these cases are accelerometer and gyroscope. So here we selected some cases which includes normal touch, accident keep, complete slip, slip till tipping point, flip and fall. For all these six positions, 20 samples each are taken. The acceleration and orientation values for each sample are stored. The data obtained can be plotted which can be filtered and the appropriate features can be extracted [13]. Based on the features extracted, classification algorithm can be implemented using machine learning so that the system can automatically classify different positions. From [14], we got to know that people do consider many aspects before placing a phone somewhere. So based on those aspects, we selected the various positions of the phone.
In section II, the methodology of how data was collected for various samples of different phone slip cases and also procedure to generate the required database is discussed. In Section III, the procedure followed to extract features from the created database is discussed. In section IV, the method used to create a database from the extracted feature is mentioned. In section V, various machine learning classification algorithms is discussed and also how these algorithms are used to classify various phone slip cases is discussed with the observations obtained after implementing the various classification algorithms is discussed and tabulated. In section VI, the final conclusion is drawn depending upon the results obtained.
In our study, we considered six phone slipping cases: normal touch keep case, accidental keep case, comp
This content is AI-processed based on open access ArXiv data.
