How to build a low-cost IoT sensor network?

"Sensor Fusion for Public Space Utilization Monitoring in a Smart City" is the best reading for Internet of Things product designers, developers and implementers. It measures the city's space utilization by designing a system, weighing various factors such as sensor selection and calibration, power supply selection, network design, data cleaning and standardization, and data processing. This method can be extended to the design of any Internet of Things network. This paper can be called a perfect case study on how to build the Internet of Things.

Billy Pik Lik Lau, Nipun Wijerathne and Chau Yuen of Singapore University of Technology and Design and Benny Kai Kiat Ng of Curtin University in Australia pointed out that the most interesting point in the paper is how they match the sensors and obtain the correct resolution data to estimate Space utilization, and establish a test platform to minimize large-scale implementation problems. In order to measure space utilization (the number of people in a space in multiple time intervals), they chose sound and motion sensors and a fusion of the two. The method used in this paper can be applied to other types of sensors.

Sound sensor and video sensor

Sound sensors are more used to detect activities than video sensors. This may seem counterintuitive, but only because the human senses are more visually dominant. The camera is expensive in terms of cost, and the increase in the amount of processing data requires the use of more expensive and more powerful computers in the network. In a large-scale deployment, this will increase the cost, but will not improve the accuracy of the measurement. More computing power will increase power consumption, exceeding the power supply capacity and cost of solar cells. The cloud processing of video data requires a lot of network bandwidth and storage space, which increases costs. Finally, due to privacy issues, the deployment of cameras requires permission, which will pose a problem when deployed in Singapore, where the research is located.

Renewable Wide Area Sensor Network (RWSN) solves the problem of hardware connection power and the need to replace the battery. Renewable energy design is not necessary for testing the use of sound and motion sensors, because the hardware connection power or battery in this limited seven-node network will not be expensive. The need for large-scale deployment of this sensor network may be the reason for choosing renewable energy.

The renewable wide-area sensor network uses a low-power XBee module (IEEE 802.15.4) to connect the XBee receiver to the Raspberry Pi and send data back to the cloud storage. Researchers use Xbee repeaters to build a wireless mesh network to increase coverage. The network is powered by solar panels and battery storage, both of which are included in size, so the IoT system designer can adjust the size of solar panels and batteries to suit local conditions according to different solar beams.

According to Memes Consulting, different environmental monitoring sensor nodes include barometers, thermometers, photometers, resistive rain sensors, ultraviolet (UV) index sensors, humidity sensors, motion and noise sensors. In the IoT network, these nodes can provide calibration data to eliminate the influence of environmental conditions such as noise reading interference caused by rain.

Low-priced pyroelectric infrared (PIR) sensors, also known as passive infrared sensors, are used to detect movement. The low-cost analog sound sensor, which is basically a MEMS microphone, is used to record sound.

PIR sensor will have a lot of wrong data output during the day, especially in the afternoon. In order to eliminate these errors, the calibration module can be used for data preprocessing. The false alarms of live ground measurements on site are related to bright sunlight. The calibration module can calculate the probability of false alarms and make adjustments, and then perform normal statistics on the data.

Apply machine learning to eliminate errors

Environmental errors in environmental conditions such as rainfall can be eliminated by using unsupervised machine learning methods, allowing researchers to use clusters to find similar patterns in sound data. The cluster simply classifies similar data sets such as the sound of rain, which can then be removed from the data. Similarly, background noise can also be eliminated.

Standardization and calibration data from PIR and sound sensors are fused using an algorithm chosen by the researchers to estimate the space utilization of the seven nodes in the entire test area. The estimation is based on the comparison of field observation experience and the fusion data of seven nodes.

This paper explains the design considerations of hardware, communications, sensors, and data processing for building a low-cost and accurate IoT system. It also explains how to eliminate false positives in the data captured by the PIR motion sensor, how to describe noise characteristics from human activities, and eliminate environmental errors such as rainfall and background noise to accurately estimate the utilization of individual nodes and the overall test platform Wait for the challenge.

The Internet of Things network needs to have the potential for large-scale implementation and proof that economic or social returns can offset design and development costs. Design and development require a multidisciplinary team with some professional skills, especially sensor engineering and advanced mathematics knowledge in subfields such as statistics and machine learning. Mathematical skills can be used to eliminate calibration data and background noise; sensing engineering skills can be used to obtain correct resolution data at low cost. Companies that are serious about building IoT sensor networks may need to hire professionals with sensor and mathematical skills.