According to MEMS Consulting, researchers have designed an Internet of Things (IoT) network that integrates sound and motion sensors to estimate the utilization of public spaces. These ideas can be applied to other IoT sensor networks. “Sensor Fusion for Public Space Utilization Monitoring in a Smart City” is a must-read for IoT product designers, developers, and implementers. It measures urban space utilization by designing systems that weigh factors such as sensor selection and calibration, power selection, network design, data cleaning and normalization, and data processing. This approach can be generalized to the design of any IoT network. The paper serves as a perfect case study on how to build an IoT. Billy Pik Lik Lau, Nipun Wijerathne, and Chau Yuen from Singapore University of Technology and Design, along with Benny Kai Kiat Ng from Curtin University in Australia, pointed out that the most interesting aspect of the paper is how they matched sensors to obtain data with the correct resolution to estimate space utilization and built a testing platform to minimize issues with large-scale implementation. To measure space utilization (the number of people in a space over multiple time intervals), they chose sound and motion sensors as well as a fusion of both. The methods used in the paper can also be applied to other types of sensors.Sound Sensors vs. Video SensorsSound sensors are more often used to detect activity compared to video sensors. This seems counterintuitive, but it is simply because human senses are more visually dominant. Cameras are expensive in terms of cost and require more expensive and powerful computers in the network to handle the increased data volume. In large-scale deployments, this increases costs without improving measurement accuracy. More computational power increases power consumption, exceeding the capacity and cost of solar panels. Additionally, cloud processing of video data requires significant network bandwidth and storage space, further increasing costs. Finally, deploying cameras requires permits due to privacy concerns, which can be problematic for deployment in Singapore, the research location. Renewable Wide Area Sensor Networks (RWSN) address the issues of hardware power connections and battery replacement needs. Renewable energy design is unnecessary for testing sound and motion sensors since the hardware connection power or batteries in this limited seven-node network won’t be costly. The need for large-scale deployment of this sensor network may be a reason to choose renewable energy. The RWSN uses low-power XBee modules (IEEE 802.15.4) to connect XBee receivers with Raspberry Pi, sending data back to cloud storage. The researchers built a wireless mesh network using XBee repeaters to increase coverage. The network is powered by solar panels and battery storage, with both sizes included so that IoT system designers can adjust the sizes of solar panels and batteries according to local conditions based on varying sunlight. According to MEMS Consulting, various environmental monitoring sensor nodes include barometers, thermometers, photometers, resistive rain gauges, UV index sensors, humidity sensors, motion and noise sensors. These nodes in the IoT network can provide calibration data to eliminate the effects of environmental conditions like noise readings caused by rain. Inexpensive passive infrared (PIR) sensors are used to detect motion, while inexpensive analog sound sensors, essentially MEMS microphones, are used to record sound. PIR sensors often output a lot of erroneous data during the day, especially in the afternoon. To eliminate these errors, calibration modules can be used for data preprocessing. False alarms from ground reality measurements are associated with bright sunlight, and calibration modules can calculate the probability of false alarms and make adjustments, then normalize the data statistically. Applying Machine Learning to Eliminate ErrorsEnvironmental errors due to conditions like rainfall can be eliminated by using unsupervised machine learning methods, allowing researchers to find similar patterns in sound data through clustering to remove them. Clustering simply classifies similar datasets such as the sound of rain, which can then be removed from the data. Similarly, background noise can be eliminated. Standardized and calibrated data from PIR and sound sensors is fused using algorithms chosen by the researchers to estimate the space utilization of the seven nodes in the entire test area. This estimate is based on field observation experience and the comparison of fused data from the seven nodes. The paper outlines the considerations for hardware, communication, sensors, and data processing design for building a low-cost, accurate IoT system. It also explains the challenges of eliminating false alarms from data captured by PIR motion sensors, how to describe noise characteristics from human activities, and how to eliminate environmental errors such as rainfall and background noise to accurately estimate the utilization of individual nodes and the overall test platform. IoT networks need to be capable of large-scale implementation and demonstrate the potential for economic or social returns to offset design and development costs. Design and development require a multidisciplinary team with some specialized skills, particularly in sensor engineering and advanced mathematics such as statistics and machine learning. Mathematical skills can be applied to calibrate data and eliminate background noise; sensor engineering skills can be used to acquire correct resolution data at low cost. Businesses seriously building IoT sensor networks may need to hire professionals with sensor and mathematical skills.
Source: MEMS
Editor: Song Bingjia
