2.1.5. Systematic Requirements

This subsection presents a well-defined set of systematic requirements provided by the devised architecture, considering the practical deployment of low-cost sensor networks for ANPR. These requirements are as follows:


Regarding scalability, the architecture proposed in this work provides support (i) when new low-cost sensors need to be integrated and (ii) when new physical locations need to be monitored. The integration of new sensors is carried out in the perceptual layer. Thus, this systematic requirement is guaranteed thanks to the existing independence between sensors. As mentioned before, each sensor is responsible for a single physical point. Similarly, when a new physical location needs to be added, then it is only necessary to deploy a new sensor which, in turn, will send information to the upper-level layer and will notify whether it is working properly. This is why integration is also guaranteed in terms of adding new devices when they are required. In other words, this requirement is strongly related to scalability of the devised architecture.

The systematic requirement named availability has been achieved thanks to the adopted cloud-based approach, since it is easy to incorporate multiple layers of license plate analysis so that processing errors are identified. Although the currently deployed system only uses OpenALPR, the architecture easily allows the incorporation of other license plate identification platforms that minimize potential errors. On the other hand, all processing sensors run the same software on the same hardware. Replacing a sensor implies changing its identifier and the server address that are specified in the configuration file stored in the memory stick. In other words, replacing a faulty sensor is a simple and straightforward task. This decision is related to the systematic requirement evolvability.

With respect to security, multiple methods have been considered to protect the information exchanged between the di fferent components of the architecture, ensuring its integrity. Particularly, the extension hypertext transfer protocol secure (HTTPS) has been used to guarantee a secure communication so that the information is encrypted using secure sockets layer (SSL). Finally, regarding manageability, the devised architecture aims at facilitating the deployment process of sensor networks. In fact, there is a component, named experiment module definition, which has been specifically designed to address this systematic requirement. As previously stated in Section 2.1.3, it is possible to set up experiments and adjust the configuration of the sensors in a centralized way, without having to individually modify the internal parameters of every single sensor.

#### *2.2. Low Cost Sensor Prototype Description*

## 2.2.1. Production Cost

From a hardware point of view, each low-cost sensor (€62.27) is composed of the following components (see Figure 4):


Raspberry Pi is a low-cost single board computer running open source software. The multiple versions of the board employ a Broadcom processor (ARM architecture) and a specific camera connector. Thanks to the use of this hardware, the versions of the Raspberry Pi OS (formerly called "Raspbian"), derived from the GNU/Linux distribution Debian, can be used. Thus, the development in a number of general-purpose programming languages is possible.

For the development of the sensor previously introduced, the version of the board called Pi Zero W has been used, which incorporates the Broadcom BCM2835 microprocessor. This has a single-core processor running at 1 GHz, 512 MB of RAM, a VidoCore IV graphics card, and a MicroSD card as a storage device. Based on the Pi Zero model, this version offers Wi-Fi connectivity, which allows online monitoring. In the conducted tests, the connectivity with the cloud has been done by using 3G/4G connection sensors, using the existing institutional Wi-Fi network of the University of Castilla-La Mancha whenever possible.

**Figure 4.** Prototype of the designed sensor.

The 8 megapixel Raspberry Pi Camera V2.1 features Sony's IMX219RQ image sensor with high sensitivity to harsh outdoor lighting conditions, with fixed pattern noise and smear reduction. The connection is made using the camera's serial interface port directly to the CSI-2 bus via a 15-pin flat cable. The camera automatically performs black level, white balance, and band filter calibrations, as well as automatic luminance detection (for changing conditions) of 50 Hz in hardware. In the configuration of each sensor, the resolution with which each photograph is taken can be specified, up to the maximum of 3280 × 2464 pixels.

The used Lithium battery holds a capacity of 5000 mAh, with an output of 5 V/2.1 A and a very small size and weight (100 × 33 × 31 mm, 195 g).

The installed operating system is based on Debian Buster, with kernel version 4.19. The installation image has a size of 432 MB which, once installed on the system partition, uses 1.7 GB. The current version of the prototype uses a UHS Speed Class 1(U1) microSD card, with a write speed of 10 MB/s required to record high-definition pictures in short intervals. Each 8 MP photograph may require around 4 MB in jpg format if stored at full resolution (depending on the scene complexity and lighting conditions).

In the conducted tests, each sensor made the captures with a resolution of 1024 × 720 pixels. Each image occupied an average of 412 KB, size that was reduced to 151 KB after the optimization process with capture sub-regions. The capture frequency was established to 1 image per second. This requires a disk storage of 1.4 GB for every hour of capture without optimization. Thus, with more than 14 GB available on the SD card for data, it is possible to store more than 8 h of images without optimization, and more than 24 h by defining capture sub-regions.

The 128 MB Flash Drive is used to store the processing sensor configuration parameters, such as the unique identifier of the processing sensor, the address of the web server associated with the intelligent experiment managemen<sup>t</sup> layer, and the network configuration.

For the integration of all hardware components of the system, a basic prototype has been made using 3D printing, with a size of 103 × 78 × 35 mm, and a unit cost of 1.18 (59 g of PLA of 1.75 mm).

The cost of the designed sensor is similar to some of the low-cost sensors discussed in [22]. However, the o ffered functionality can be compared to commercial systems with a significantly greater cost. Plus, the devised architecture enhances the global functionality of the sensor networks deployed from the architecture, and this is a major improvement regarding existing work in the literature.
