Occupancy monitoring plays a key role in building automation, health, safety, and security. Although developers can piece together suitable people counting solutions from available components, and develop the appropriate algorithms, this can be time-consuming and costly. Amid heightened expectations for faster delivery of solutions with more sophisticated and up-to-date capabilities and features, including support for social distancing requirements, a simpler, faster approach is required.
This article discusses occupancy monitoring and why it has become such a critical feature. It then introduces and describes how to get started with a comprehensive, end-to-end people-counting kit from Analog Devices. Using the kit, designers can meet the diverse requirements for an expanding list of sophisticated applications based on occupancy monitoring functionality.
Why occupancy monitoring matters
The ability to monitor the number of individuals, their location, and their movement within a building is finding an expanding role in multiple applications. Within automated building management systems (BMS), the ability to track room utilization and occupant movements remains fundamental to realizing the full benefit of offices, meeting rooms, and other common areas. During pandemic surges, this capability helps ensure that occupants can maintain safe separation in indoor spaces.
Even as individuals return to office buildings, the ability to monitor room occupancy helps companies limit energy wasted on the typically high number of unused building spaces. Office occupancy rates that were already down to about 68% in 2019 [a] plummeted during the pandemic, returning to only about 32% in mid-2021 [b].
Beyond optimizing the use of building spaces and aiding social distancing, however, an active measure of occupancy has become essential to throttle escalating energy consumption. According to the World Green Building Council , buildings and construction contribute 39% of all carbon emissions globally. More specifically, energy used to light, heat, and cool buildings account for 28% of worldwide carbon emissions. (The remaining 11% relates to building lifecycle carbon costs of materials and construction.)
After remaining flat for most of the last decade, building-related carbon emissions rose to an all-time high in 2019 due to increased energy demand driven by more extreme weather. In fact, 2019 proved to be the hottest year on record since 2016 when global weather patterns and rising global temperatures combined in a “perfect storm” of unusually warm weather.
This trend for warmer weather has continued, with 2020 proving to be warmer than 2019. As a result, the three warmest years on record to date now include 2016 (1st), 2020 (2nd), and 2019 (3rd), according to the US National Oceanic and Atmospheric Administration (NOAA) . The trend continues with July 2021 recorded as the hottest month ever recorded worldwide . With the four months preceding July each ranked in the top 10 warmest months ever recorded , NOAA anticipates that 2021 will likely become one of the top 10 warmest years ever recorded globally.
Globally, national strategies to reduce climate-affecting carbon emissions view more efficient building energy utilization as central to their planning. For individual companies, reduced energy consumption offers direct benefits to their bottom line, as well as their employees’ well-being.
Despite the growing importance of basic occupancy data in minimizing energy utilization, most companies rely on access badge swipe data or visual observation—neither of which can provide the precise, updated information on room utilization needed for effective building energy management. A more effective means of occupancy sensing is required.
Implementing an occupancy sensing solution
The design and implementation of an automated occupancy sensing solution requires expertise in multiple areas in order to combine sensors, low-power processors, and connectivity with accurate people-counting algorithms into full applications, able to respond instantly as people enter and leave indoor spaces. This takes time and resources to develop and support. Analog Devices offers a simpler route: the ADSW4000 EagleEye, a complete 2D vision sensor-based, low-power, low-bandwidth drop-in platform designed specifically to provide updated data to optimize space utilization and minimize energy consumption.
The kit comprises the Analog Devices’ proprietary People Count algorithm running on a member of Analog Devices’ ADSP-BF707 series of Blackfin digital signal processors (DSPs). The ADSW4000 EagleEye delivers utilization data for separate indoor spaces, allowing companies to balance office space utilization and energy consumption for maximum utility.
Because it performs its image analysis and people counting tasks solely on the Blackfin processor, the EagleEye algorithm ensures that all images remain on the ADSW4000, so no personally identifiable information ever leaves the platform, conforming with a growing body of worldwide privacy regulations. In fact, results generated by the Blackfin processor are limited to a data package containing the number of people in a monitored region of interest (ROI), their x,y location in that region, and whether or not they are on the move.
To help speed the development of high-level occupancy monitoring applications, Analog Devices integrates its ADSW4000 EagleEye People Count platform in its EVAL-ADSW4000KTZ EagleEye trial kit. Serving as a complete sensor-to-cloud turnkey implementation of its EagleEye algorithm, the trial kit enables users to immediately deploy occupancy monitoring using the available app and cloud-based online dashboard. Alternatively, the kit can serve as the foundation of custom systems, allowing developers to focus on their higher-level applications rather than the details of implementing their own people counting methods.
Individual subsystems accelerate implementation
The EagleEye trial kit comprises a pair of subsystems, using one subsystem based on the Blackfin DSP to generate people counting data, and a separate subsystem based on Analog Devices’ ADuCM4050 microcontroller unit (MCU) to handle connectivity and higher-level application functionality (Figure 1). As mentioned earlier, the critical people counting functionality lies in the trial kit’s EagleEye DSP subsystem running the ADSW4000 EagleEye algorithm.
Figure 1: In Analog Devices’ EagleEye trial kit, a DSP subsystem acquires and processes images using the ADSW4000 EagleEye PeopleCount algorithm running on a member of Analog Devices’ ADSP-BF707 Blackfin DSP series. (Image source: Analog Devices)
For image acquisition from the region of interest, the subsystem uses a 2D vision sensing module based on onsemi’s ASX340AT3C00XPED0-DPBR CMOS digital image system-on-chip (SoC) combined with an infrared (IR) filter. Working with Analog Devices’ EagleEye framework services, the EagleEye PeopleCount ADSW4000 algorithm runs on the ADSP-BF707 Blackfin DSP using ISSI’s IS25LP512M 512 megabit (Mbit) serial flash memory and Micron Technology’s MT46H64M16LF 1 gigabit (Gbit) low-power double data rate (DDR) synchronous dynamic random access memory (SDRAM).
In this subsystem, the ADSP-BF707 Blackfin DSP is well suited to handling the complex image acquisition and processing tasks required for people counting. Its signal processing pipeline includes multiple hardware multiply-accumulate (MAC) units along with single instruction, multiple-data (SIMD) capabilities.
Running on the ADSP-BF707 Blackfin processor, the ADSW4000 ADI EagleEye PeopleCount algorithm achieves up to a 90% accurate count within the target area. Just as important, the subsystem returns the results quickly. For example, the subsystem needs only 300 milliseconds (ms) from the time a person enters an ROI to identify that it has switched from a vacant to an occupied state. The time required to identify a change in ROI state from occupied to vacant is user configurable, with a default setting of five minutes.
Latency is similarly low for generated people count and location data. The algorithm provides updated people count and location data within 1.5 seconds after an individual moves into a zone defined by the user during commissioning. After detecting an individual, the algorithm needs only 113 ms to provide updated count and location data.
As noted above, Analog Devices’ EagleEye platform does not transmit any captured images. Instead, the DSP uses its universal asynchronous receiver-transmitter (UART) port in push mode to transmit occupancy metadata. Transmitted in JSON format, this metadata packet includes occupancy state (occupied or vacant), people count, people location as x,y coordinates, along with other data (Table 1).
Table 1: Analog Devices’ EagleEye algorithm maintains users’ privacy by not transmitting personally identifiable information, but instead generating a package that includes the metadata listed here. (Table source: Analog Devices)
Downstream of the DSP subsystem, the ADuCM4050 MCU subsystem runs in the AWS FreeRTOS environment, supporting the high-level EagleEye application and connectivity services required for sensor commissioning and communication with Analog Devices’ associated cloud-based service (Figure 2).
The 32-bit ADuCM4050 MCU offers a comprehensive processing environment for Industrial Internet of Things (IIoT) applications like Analog Devices’ EagleEye. To support complex industrial application workloads, the ADuCM4050 builds on an Arm® Cortex®-M4F 52 megahertz (MHz) processor core with integrated floating point unit (FPU), memory protection unit (MPU), hardware cryptographic accelerator, and protected key storage.
Figure 2: Based on Analog Devices’ ADuCM4050, the EagleEye trial kit’s MCU subsystem supports the higher level IIoT application and provides connectivity services locally and between the kit and the cloud or other building management systems. (Image source: Analog Devices)
A set of integrated power management features including multiple power modes and clock gating capabilities enable the device to achieve low power execution. As a result, the MCU requires requiring only 41 microamps per megahertz (μA/MHz) (typical) in active mode and 0.65 μA (typical) in hibernation mode. During inactive periods, the processor consumes only 0.20 μA (typical) in its fast-wake-up shutdown mode or only 50 nanoamps (nA) in full shutdown mode.
How to quickly begin people counting
In the trial kit, Analog Devices combines the DSP and MCU subsystems with a camera sensor, a lens, LEDs, and buttons in a compact package (Figure 3).
Figure 3: Designed for quick deployment, the 2D vision sensor unit in Analog Devices’ EagleEye trial kit can be easily mounted above a region of interest for people counting. (Image source: Analog Devices)
Developers can quickly deploy people counting by simply mounting the sensor unit in a room or indoor space, directly above a region of interest. The sensor can use power from a variety of sources. Users can run a wire to the unit’s DC connector to supply a 5.5 to 36 volt DC source, or power it by supplying a USB power source using a micro USB cable, or active USB extension for distances beyond 1 meter (m).
After mounting the sensor unit, users can visually confirm the sensor positioning and desired field of view (FOV) using the companion EagleEye PeopleCount app available on the Apple App Store for iOS tablets, or on Google Play for Android tablets (Figure 4).
Figure 4: Analog Devices’ EagleEye PeopleCount app allows easy confirmation of the sensor unit placement prior to commissioning. (Image source: Analog Devices)
After users verify the sensor FOV, they proceed with the brief device commissioning process. During commissioning and later during operation, users can observe the DSP and MCU LEDs built into the sensor unit to monitor the current status of the respective subsystems (Table 2).
Table 2: Separate LEDs built into the Analog Devices’ EagleEye trial kit sensor unit provide a continuous indication of state of the DSP and MCU subsystems. (Table source: Analog Devices)
The app walks users through the few steps required for sensor commissioning. In this process, users indicate which areas the algorithm should monitor within the FOV by marking a series of inclusive masks, such as the floor mask (Figure 5, left). Areas to exclude are just as important for accurate counts. During the commissioning process, the companion app allows users to specify different exclusion masks, for example, windows and display screens (Figure 5, right).
Figure 5: During commissioning, users employ the companion app to identify areas that the EagleEye PeopleCount algorithm should examine or ignore, using inclusive masks such as the floor mask (left), and exclusive masks (right) for windows or other areas that degrade people counting accuracy. (Image source: Analog Devices)
Once mounted and commissioned, the sensor unit begins transmitting its metadata to Analog Devices’ cloud. By logging into the cloud using credentials provided during registration, users can examine a series of graphical representations of occupancy (Figure 6).
Figure 6: After mounting and commissioning the Analog Devices’ EagleEye trial kit sensor unit, users can log in to an online dashboard in Analog Devices’ cloud to view real-time occupancy data. (Image source: Analog Devices)
Analog Devices’ EagleEye PeopleCount technology platform can be embedded in custom designs built with the appropriate Blackfin processor and suitable external flash memory. Analog Devices also makes the EagleEye software package available for registered trial kit customers. For the downstream MCU subsystem, developers can provide additional functionality including more sensors, using any system platform design able to run the EagleEye sensory interface and provide the required connectivity. For developers looking to quickly employ people counting in their building management systems, however, the Analog Devices EagleEye trial kit offers a turnkey sensor-to-cloud solution.
As companies pay the price of significant building energy consumption from office lighting, heating, and cooling, effective resource management of often vacant office spaces is driving a need for more accurate occupancy data. Based on a proprietary algorithm running on a low-power digital signal processor, the ADSW4000KTZ Trial Kit provides a comprehensive sensor-to-cloud platform for evaluating and deploying occupancy monitoring, able to provide the real-time, room-level data on occupancy needed for more effective building energy management.