**7. Conclusions**

A novel methodology to measure traffic volumes without supervision was developed in the present study. An autoencoder was trained to recognize the presence of vehicles on a cross section of a roadway based solely on pixel intensities. The proposed methodology requires no human effort to tag images with labels. The performance of the proposed model trained on synthesized images approximated that of a YOLO-based model that was fine-tuned with the extra labeling of images. When considering the human effort required for the labeling task, the proposed methodology seems more promising for use in the field.

The proposed methodology, however, demonstrated a critical drawback wherein vehicle types could not be distinguished. It is possible to approximately classify vehicle types if a constant speed is assumed for all moving vehicles. This assumption is acceptable when the detection line is placed on the stop line of an intersection approach. Speed estimation that depends on spot detectors usually adopts such an assumption. Further study is expected to develop a traffic volume counter that could classify each vehicle type.

**Author Contributions:** Conceptualization, K.S.; Data curation, J.S.; Formal analysis, S.R.; Writing—original draft, K.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by the Chung-Ang University Research Scholarship Grants in 2019 and also by the Korea Agency for Infrastructure Technology Advancement (KAIA) gran<sup>t</sup> funded by the Ministry of Land, Infrastructure, and Transport (grant number 19TLRP-B148659-02).

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