Among the challenges faced by the healthcare sector is a population that is growing older. The elderly population is expected to grow significantly over the next 20 years. Having an independent lifestyle is highly desired by elderly people, but independence for older adults often comes with high risks. Many smart home technologies have been developed to track and monitor activities of the elderly at home and assist their independent living. Buildings and urban environments fitted out with sensor networks offer the elderly the chance to retain their independence for longer. Wearable sensor technologies can also play a major role.
Human behavior analysis and activity recognition is an integral part of today’s home healthcare systems. It is essential to have reliable and accurate monitoring as well as real-time actuation when needed. Activities of daily living, such as cooking, sleeping, and cleaning, are good indicators of the physical capabilities of elderly or sick. Therefore, a system that automatically recognizes these activities can allow seamless health monitoring and provides an objective measure for medical staff. Also such a system should be able to detect any anomalies such as a sudden fall and provide for an immediate actuation. An activity monitoring system is therefore a crucial step in the future health applications.
This article introduces an integrated home health monitoring system that includes both a vision-based activity monitoring system and a vital sign monitoring system. These two systems monitor the activities of an individual and enable the individual to monitor his or her vital signs while performing that activity. The integration of wearable healthcare technology and embedded vision technology is the key to achieving a true home health monitoring system.
Until now, activity monitoring market has been largely dominated by video surveillance technology. However, with the shift of this activity monitoring to home environments, video cannot offer the necessary privacy and cannot support the amount of data that must be transmitted. By contrast, an embedded vision sensing platform performs real-time processing at the edge node with the system output being only useful telemetry and processed data. Because only the telemetry data are transmitted, the system saves more than 90 percent of the bandwidth requirements. An activity monitoring embedded vision system detects people, tracks them as they move, and recognizes their postures and activities of interest. The architecture of an embedded sensing platform typically consists of the following:
An optics system (a cmos sensor plus the lens) – This system does the image capture. The optics configuration needs to be defined on the basis of the field of view and the configuration of the system as well as the geometry of the room. Sometimes the cmos sensor might do some image preprocessing, thus reducing the load of processing for the embedded processor, which, in turn, could reduce the power consumption of the system if correctly duty-cycled.
The processing system – The processor is the heart of this system. It is now expected to do a lot more controlling, sensing, and interfacing while consuming very little power and area. The processor in an embedded system platform runs the image processing algorithm on the image captured by the optics system. The output of the system consists only of telemetry data. In a home health scenario, the output could be about the activity of the person, i.e., whether he or she is sleeping or cleaning or has fallen down.
Connectivity – The connection could be either wired or wireless in an embedded vision-based system. However, in a home environment, its likely to be wireless connectivity. Because the output is only telemetry data and not raw video data, the payload to transmit is greatly reduced. This data is then transmitted to a cloud and made available in real-time in the form of an app to a nurse or guardian.
Cloud/data analytics – The analytics form the back end of the system. Not only does the cloud infrastructure provide a real-time access to the data in the form of an app, but it also can run a background data analytics algorithm to identify trends in context with home activities (see Figure 1).
System Design Considerations and Major Challenges
Reliability – It’s highly essential that an activity monitoring system provides the activity information in the most reliable, secure, and accurate fashion. Also, in case of an emergency, the system must be able to accurately detect an emergency environment and set out an alarm with a focus on reduced false-alarm generation.
Latency – This is an immediate response/actuation/alarm generated by activity monitoring systems that defines the potential of the security system. The basic functionality of monitoring activity — sleeping, walking, cleaning, or an emergency — should be reported instantaneously to minimize time between the occurrence and reporting.
Tamperproof – An activity monitoring system needs to be as tamperproof as possible. Tampering can occur at any stage of the system: end node, wireless/ wired connectivity, or at the data control and analytics end of the cycle. Breach of building automation systems or networks is of great concern in home monitoring systems.
Embedded Vision Sensing Platform
The Blackfin low power imaging platform (BLIP) is a low-cost, low power, high-performance embedded vision sensing platform that can run an array of real-time sensing and image-processing algorithms.
Accurate, compact, and low-power vital sign measurements – ADI demonstrated the technology at the CES show in January 2016, The measurement of vital signs presented included heart rate and activity, and it was shown by a watch on the wrist (see Figure 2).
The vital signs measurements (VSM) included heart rate and activity. A modular architecture consists of a motherboard embedding ADI’s Cortex M3 microcontroller, the ADuCM302x, and a 2.4 GHz radio transceiver allowing it to send VSM data with the Google Thread protocol. The daughter board contains analog front end optics surrounded by three green LEDs and a photodiode, as a low-power three-axis accelerometer. These two devices are synchronized with one another to compensate for the movement of the person more efficiently, using an algorithm designed for this purpose.
The ADPD103 is a photometric front end that works using reflective optical measurement by sending an 8–250 mA current through its LED drivers to illuminate the external LEDs of the component. These LEDs illuminate the skin. Using a reflected measurement through the photodiode, the signal is acquired by the front end, and is then amplified, filtered, integrated, and converted by a 14-bit ADC before being transmitted to a host via an I²C interface.
Having external LED and photodiodes allows selection of the number and color of the LEDs, their current strength, and the optimal spacing of the LEDs with the photodiode to maximize the modulation index, which sets the AC/DC ratio and thus the quality of the reflected signal. An external configuration also lets the designer choose the size of the photodiode (when the latter is wider, the modulation index will be higher), and possibly add an ultralow noise and low-power current amplifier to it.
For measurement of the heart rate on the wrist, green LEDs were used because the hemoglobin absorption is highest for the 500–600 nm wavelengths. When a heart beats, the blood flow in the wrist — and the absorption of green light — is superior. Between beats, it decreases. By flashing green LEDs hundreds of times per second, the ADPD103 can calculate the number of times the heart beats every minute. The green LED to the photodiode should be spaced about 3 cm or more conveniently to increase the modulation index (see Figure 3).