**4. Conclusions**

This paper presents a proposal for deployment of a low-cost sensor network for automated vehicles plate recognition in a pilot project in Ciudad Real (Spain). For this, three main tools were needed: (1) the architecture to deploy the sensors, (2) a low-cost sensor prototype, and (3) a methodology to decide the best location for the sensor.

Regarding the deployment of sensors and the sensors themself, one of main features to highlight is that the total cost is very low in terms of the following elements:


In addition, the deployed sensor is completely decoupled from the specific license plate identification platform used. This allows to change the platform if the user found any other better. In particular, the used platform identifies besides the license plate number, the vehicle's manufacturer, model, and color data. This information can be used in the overall analysis of tra ffic flows with a view to reducing possible errors in the identification of the number plate and will be developed in the future by the authors.

The third tool used in this paper is a methodology to determine the location in the tra ffic network of the designed sensors. To this, we have proposed the use of two algorithms which aim to achieve a good enough quality of the tra ffic flow estimation to be done (in terms of low RMARE value) with the ANPR data collected by the sensors.

The model was applied to the tra ffic network of a pilot project considering a deployment of 30 sensors analyzing whether or not to install the proposed sensors on some links due to the di fficulty of its installation. The results were very positive since the expected quality of the estimation results is very similar to those obtained when allowing the sensor to be located in any link. The main advantage is that avoiding those conflictive links we expect a reduction obtaining errors of reading vehicle plates.

The influence of other parameters of the model were also analyzed such as the number of routes used as reference and the degree of network simplification. The analysis of the results shows that considering a greater number of reference routes, represented by means of the parameter *k*, leads to a better estimation of the flows in terms of achieving a smaller *RMARE*. However, a high value for *k* would imply working with a network with a large number of routes, which would have a high computational cost. In reference to network simplification, a medium–low degree of network simplification leads to a good performance of the methodology in terms of the error obtained in the estimation step.

**Author Contributions:** Conceptualization, all authors; methodology: F.Á.-B. and S.S.-C.; location model: F.Á.-B., S.S.-C., A.R., and I.G.; architecture of sensor deployment: D.V., C.G.-M., and S.S.-C.; validation, all authors; formal analysis, F.Á.-B. and S.S.-C.; investigation, F.Á.-B. and D.V.; resources, A.R. and I.G.; data curation, S.S.-C. and C.G.-M.; writing—review and editing, F.Á.-B., S.S.-C., D.V., and C.G.-M.; project administration, A.R. and S.S.-C.; and funding acquisition, A.R. and S.S.-C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work is funded thanks to the financial support of the Spanish Ministry of Economy and Competitiveness in relation to project TRA2016-80721-R (AEI/FEDER, UE).

**Acknowledgments:** The authors acknowledge Miguel Carrión (University of Castilla-La Mancha) and the university's technical staff for providing computer resources. We also want to acknowledge the editor and the two anonymous reviewers whose valuable comments have contributed to improve this work.

**Conflicts of Interest:** The authors declare no conflict of interest.
