08-10-2016, 10:30 AM
[attachment=72706]
Abstract— In this paper we propose an algorithm and anarchitecture for the fall detection and wide area rescue system based on a smart phone and 3G networks. The angles acquired by the electronic compass or generally known as e-compassand the tri-axial accelerometer’s waveform sequence on the smart phone are used as the system inputs to realize the fall detection algorithm. Thesignals acquired are used to generate an ordered sequence and then the proposed cascade classifier examines them in a sequential manner forrecognition. When the corresponding feature is verified at present state, it can proceed to next state; or else, the system will reset to the initial state and wait for the occurrence of another sequence. When a fall event is detected, the position of the user can be obtained by the global positioning system (GPS) or the assisted GPS (A-GPS), and is sent to the rescue center using the 3G network so that the user can get medical assistance immediately. In the proposed architecture, the power consumption issue and computational burden on the smart phone can be reduced. Moreover, as we will see in the experiment that a distinguished fall accident detection accuracy up to 99.75% on the specificity and 92% on the sensitivity.
INTRODUCTION
The major cause of injury to the elderly in recent years is due to the fall accident. The system is proposed to protect the elderly from the injury of fall accident events or to give an immediate assistance to the elderly after the occurrence of a fall accident event. Among all the currently proposed algorithms, the fall detection system can be roughly divided into two categories, namely, environmental monitoring-based and wearable sensor-based systems.
As for the environmental monitoring-based systems, typically used sensors such as cameras , acoustic sensors, radar and infrared sensors , pressure sensors , or accelerometer for vibration detection are placed in a predefined space or environment to monitor the activities of the elderly as well as the occurrence of a fall accident event. Compared to the type of wearable sensor-based system, the environmental monitoring-based fall detection system is more comfortable to the elderly since there is no need of wearing any module. However, the environmental monitoring-based system can only function in a predefined environment where it is installed. Moreover, the protection of the private matters for the elderly is another problem and contention is usually discussed with the environmental monitoring-based system .
Wearable sensor-based fall detection systems, some of which employ the use of a tri-axial accelerometer as the system input, while most of them apply the use of multiple sensors . Among the algorithms that use multiple sensors, multiple tri-axial accelerometers or a tri-axial accelerometer in conjunction with a gyroscope is usually applied. In certain multiple sensor-based systems, even the atmospheric air pressure (or barometric pressure) sensor or a surface electromyography sensor are used to assist the tri-axial accelerometer in discriminating the posture as well as the motion of the elderly.
Unlike the environmental monitoring-based systems that can function only in a predefined space, the wearable sensor-based fall detection systems can function in a larger area. However, most of the wearable sensor-based fall detection systems are made of a self-designed circuit module that should be placed and fastened around certain position, e.g., the chest or the waist, of the user. Therefore, the necessity of wearing an additional sensor module can lead to certain degree of inconvenience. In addition, how to address the current position of the elderly when a fall accident event occurs is a problem to be solved in wearable sensor-based fall detection systems. Moreover, the power consumption burden is another concern.
In order to monitor the status of the elderly and to locate the user’s position when a fall accident event is detected, some of the algorithms propose the use of a ZigBee sensor network to communicate between the fall detector that is with the elderly and the monitoring center.
Whereas to locate the user’s current position with the use of a Zigbee sensor network is not suitable for wide area applications.
We propose in this paper a pocket-based fall accident detector that uses a smart phone as the platform of the system. The tri-axial accelerometer as well as the electronic compass will be used as the sensors to generate primitive input signals.
The global positioning system (GPS) or the A-GPS will be used to acquire the user’s current position, and the longitude and latitude will be sent to the coordination center using the third generation (3G) communication network if a fall event is detected. This way, the coordination center can know the user’sposition exactly. Moreover, the detailed position can also be shown on the screen of the coordination center with electronic maps, e.g., Google map, so that the user can get medical assistance immediately. Meanwhile, the system will send out a loud sound as a warning signal so that people nearby can notice this fall accident event and provide assistance to the user immediately. With the proposed algorithm and architecture, the computational and the power consumption burden can be quite alleviated since we check each fall accident feature sequentially and reset to the initial state once any one of the feature in the state machine is not verified.
II. RELATED WORKS
M.N. Nyan, Tay F.E, Murugasu. E [1] proposed a wearable system for pre-impact fall detection. Here they use a three Gyroscope sensors which is placed in front of the waist, the sternum and the under arm to detect the fall event. This system has 100% specificityi.e. it does not shows any false alarms and also the sensitivity of this system is up to 95.2%. In spite of the accuracy these wearable sensors causes discomfort to the elderly people. G.N. Swathi and M. Amarnadh [2] proposed a Threshold based fall detection and prediction method using tri axial accelerometer. In this paper, daily activities are monitored using tri-axial accelerometer. The threshold is initially set and if the value from the sensor exceeds the threshold value, the information will be sent to a registered phone number. This system predicts the fall event 200-400 ms ahead the occurrence of the collision. Even though the fall event is detected, knowing the exact location of the elderly becomes tedious.Sung-Ihk Yang and Sung-Bae Cho [3] proposed a system for Recognizing Human Activities from Accelerometer and Physiological Sensors. Herethey use physiological sensor to detect the user’s motion in addition with 2-axis accelerometer. Using physiological signals with accelerometer
is more efficient in recognizing activities. But,the total accuracy of this system appears to be only 74.4%, which is not sufficient for effective fall detection system. Rutuja S. Shewale and Jyoti M. Pingalkar [4] proposed a system for Account of Techniques Identifying In-House Falling of Elderly People. It means that a non-wearable fall detection technology which uses camera and floor sensor to detect the fall of elderly people. This system is fast, comfortable and highly accurate. Privacy of the individual is concerned as it includes usage of cameras. P.Bharathi and G.Revathi [5] proposed a system called A Dynamic Evidential Fall Monitoring and Detection System for Elder Persons. This system is designed to detect the fall using micro-controller technology, the accelerometer and GSM modem to send out SMS to the care taker. The system uses wireless sensors which is also a cost efficient one. Though the system uses wireless sensors to send an alert message that a fall event has occurred, finding the location of the event is another tedious task.
III. SYSTEM OVERVIEW
The proposed fall accident detection and rescue system’s architecture is mainly composed of three blocks: the smart phone-based pocket fall accident detector, the coordination center and the rescue center or the first-aid stations. The motion activities for the elderly are obtained by using tri-axial accelerometer and the e-compass. The e-compass is used to acquire the tilt angle, i.e., pitch, of the smart phone. We can obtain the pitch angle of the smart phone by using a gyroscope that provides the angular acceleration information of the smart phone. But the tilt angle i.e., pitch of the smart phone can be determined by using the e-compass in conjunction with the tri-axial accelerometer. The e-compass is used for the estimation of pitch angle so that the proposed algorithm can be applied for most of the smart phone systems.
A loud sound as a signal is send as a warning once a fall event is detected and then the current position i.e., the latitude and longitude, of the elderly will be transmitted to the coordination center via the 3G network. To receive the current position and important personal information of the elderly, we propose a coordination center that composed of an emergency signal handling program module. After receiving the latitude and longitude, it can then be integrated and displayed with an electronic map, e.g., Google map.
SIGNAL ACQUISITION AND FEATURES SELECTION
Smart devices are equipped with certain kinds of detectors and the orientation of the device can be recognized by using the G-Sensor (also known as tri-axial accelerometer), the electronic compass, or the gyroscope. The tri-axial accelerometer (G-Sensor) and the electronic compass are used as the major sensors for input signal acquisition and generation in the proposed system because of their availability.
V. CONCLUSION
To detect fall and accidents, a smart phone-based pocket fall accident detection system is proposed here. Once the corresponding feature is verified by the current state, it can proceed to next state; otherwise, the system resets to the initial state and waiting for the appearance of another feature sequence. The early states are composed of simple and important features to speed up the classification process that allow a large number of negative samples to be quickly excluded from being regarded as a fall event. The computational and power consumption burden of the system can be reduced using the proposed system. By using the proposed cascaded classifier with SVM, the performance of the system is up to 99.75% on the specificity and 92% on the sensitivity can be acquired when a set of 450 test activities are evaluated.
Abstract— In this paper we propose an algorithm and anarchitecture for the fall detection and wide area rescue system based on a smart phone and 3G networks. The angles acquired by the electronic compass or generally known as e-compassand the tri-axial accelerometer’s waveform sequence on the smart phone are used as the system inputs to realize the fall detection algorithm. Thesignals acquired are used to generate an ordered sequence and then the proposed cascade classifier examines them in a sequential manner forrecognition. When the corresponding feature is verified at present state, it can proceed to next state; or else, the system will reset to the initial state and wait for the occurrence of another sequence. When a fall event is detected, the position of the user can be obtained by the global positioning system (GPS) or the assisted GPS (A-GPS), and is sent to the rescue center using the 3G network so that the user can get medical assistance immediately. In the proposed architecture, the power consumption issue and computational burden on the smart phone can be reduced. Moreover, as we will see in the experiment that a distinguished fall accident detection accuracy up to 99.75% on the specificity and 92% on the sensitivity.
INTRODUCTION
The major cause of injury to the elderly in recent years is due to the fall accident. The system is proposed to protect the elderly from the injury of fall accident events or to give an immediate assistance to the elderly after the occurrence of a fall accident event. Among all the currently proposed algorithms, the fall detection system can be roughly divided into two categories, namely, environmental monitoring-based and wearable sensor-based systems.
As for the environmental monitoring-based systems, typically used sensors such as cameras , acoustic sensors, radar and infrared sensors , pressure sensors , or accelerometer for vibration detection are placed in a predefined space or environment to monitor the activities of the elderly as well as the occurrence of a fall accident event. Compared to the type of wearable sensor-based system, the environmental monitoring-based fall detection system is more comfortable to the elderly since there is no need of wearing any module. However, the environmental monitoring-based system can only function in a predefined environment where it is installed. Moreover, the protection of the private matters for the elderly is another problem and contention is usually discussed with the environmental monitoring-based system .
Wearable sensor-based fall detection systems, some of which employ the use of a tri-axial accelerometer as the system input, while most of them apply the use of multiple sensors . Among the algorithms that use multiple sensors, multiple tri-axial accelerometers or a tri-axial accelerometer in conjunction with a gyroscope is usually applied. In certain multiple sensor-based systems, even the atmospheric air pressure (or barometric pressure) sensor or a surface electromyography sensor are used to assist the tri-axial accelerometer in discriminating the posture as well as the motion of the elderly.
Unlike the environmental monitoring-based systems that can function only in a predefined space, the wearable sensor-based fall detection systems can function in a larger area. However, most of the wearable sensor-based fall detection systems are made of a self-designed circuit module that should be placed and fastened around certain position, e.g., the chest or the waist, of the user. Therefore, the necessity of wearing an additional sensor module can lead to certain degree of inconvenience. In addition, how to address the current position of the elderly when a fall accident event occurs is a problem to be solved in wearable sensor-based fall detection systems. Moreover, the power consumption burden is another concern.
In order to monitor the status of the elderly and to locate the user’s position when a fall accident event is detected, some of the algorithms propose the use of a ZigBee sensor network to communicate between the fall detector that is with the elderly and the monitoring center.
Whereas to locate the user’s current position with the use of a Zigbee sensor network is not suitable for wide area applications.
We propose in this paper a pocket-based fall accident detector that uses a smart phone as the platform of the system. The tri-axial accelerometer as well as the electronic compass will be used as the sensors to generate primitive input signals.
The global positioning system (GPS) or the A-GPS will be used to acquire the user’s current position, and the longitude and latitude will be sent to the coordination center using the third generation (3G) communication network if a fall event is detected. This way, the coordination center can know the user’sposition exactly. Moreover, the detailed position can also be shown on the screen of the coordination center with electronic maps, e.g., Google map, so that the user can get medical assistance immediately. Meanwhile, the system will send out a loud sound as a warning signal so that people nearby can notice this fall accident event and provide assistance to the user immediately. With the proposed algorithm and architecture, the computational and the power consumption burden can be quite alleviated since we check each fall accident feature sequentially and reset to the initial state once any one of the feature in the state machine is not verified.
II. RELATED WORKS
M.N. Nyan, Tay F.E, Murugasu. E [1] proposed a wearable system for pre-impact fall detection. Here they use a three Gyroscope sensors which is placed in front of the waist, the sternum and the under arm to detect the fall event. This system has 100% specificityi.e. it does not shows any false alarms and also the sensitivity of this system is up to 95.2%. In spite of the accuracy these wearable sensors causes discomfort to the elderly people. G.N. Swathi and M. Amarnadh [2] proposed a Threshold based fall detection and prediction method using tri axial accelerometer. In this paper, daily activities are monitored using tri-axial accelerometer. The threshold is initially set and if the value from the sensor exceeds the threshold value, the information will be sent to a registered phone number. This system predicts the fall event 200-400 ms ahead the occurrence of the collision. Even though the fall event is detected, knowing the exact location of the elderly becomes tedious.Sung-Ihk Yang and Sung-Bae Cho [3] proposed a system for Recognizing Human Activities from Accelerometer and Physiological Sensors. Herethey use physiological sensor to detect the user’s motion in addition with 2-axis accelerometer. Using physiological signals with accelerometer
is more efficient in recognizing activities. But,the total accuracy of this system appears to be only 74.4%, which is not sufficient for effective fall detection system. Rutuja S. Shewale and Jyoti M. Pingalkar [4] proposed a system for Account of Techniques Identifying In-House Falling of Elderly People. It means that a non-wearable fall detection technology which uses camera and floor sensor to detect the fall of elderly people. This system is fast, comfortable and highly accurate. Privacy of the individual is concerned as it includes usage of cameras. P.Bharathi and G.Revathi [5] proposed a system called A Dynamic Evidential Fall Monitoring and Detection System for Elder Persons. This system is designed to detect the fall using micro-controller technology, the accelerometer and GSM modem to send out SMS to the care taker. The system uses wireless sensors which is also a cost efficient one. Though the system uses wireless sensors to send an alert message that a fall event has occurred, finding the location of the event is another tedious task.
III. SYSTEM OVERVIEW
The proposed fall accident detection and rescue system’s architecture is mainly composed of three blocks: the smart phone-based pocket fall accident detector, the coordination center and the rescue center or the first-aid stations. The motion activities for the elderly are obtained by using tri-axial accelerometer and the e-compass. The e-compass is used to acquire the tilt angle, i.e., pitch, of the smart phone. We can obtain the pitch angle of the smart phone by using a gyroscope that provides the angular acceleration information of the smart phone. But the tilt angle i.e., pitch of the smart phone can be determined by using the e-compass in conjunction with the tri-axial accelerometer. The e-compass is used for the estimation of pitch angle so that the proposed algorithm can be applied for most of the smart phone systems.
A loud sound as a signal is send as a warning once a fall event is detected and then the current position i.e., the latitude and longitude, of the elderly will be transmitted to the coordination center via the 3G network. To receive the current position and important personal information of the elderly, we propose a coordination center that composed of an emergency signal handling program module. After receiving the latitude and longitude, it can then be integrated and displayed with an electronic map, e.g., Google map.
SIGNAL ACQUISITION AND FEATURES SELECTION
Smart devices are equipped with certain kinds of detectors and the orientation of the device can be recognized by using the G-Sensor (also known as tri-axial accelerometer), the electronic compass, or the gyroscope. The tri-axial accelerometer (G-Sensor) and the electronic compass are used as the major sensors for input signal acquisition and generation in the proposed system because of their availability.
V. CONCLUSION
To detect fall and accidents, a smart phone-based pocket fall accident detection system is proposed here. Once the corresponding feature is verified by the current state, it can proceed to next state; otherwise, the system resets to the initial state and waiting for the appearance of another feature sequence. The early states are composed of simple and important features to speed up the classification process that allow a large number of negative samples to be quickly excluded from being regarded as a fall event. The computational and power consumption burden of the system can be reduced using the proposed system. By using the proposed cascaded classifier with SVM, the performance of the system is up to 99.75% on the specificity and 92% on the sensitivity can be acquired when a set of 450 test activities are evaluated.