*3.1. Survey Material and Equipment*

Two types of equipment were used to perform the experiment described in this paper to acquire reliable naturalistic vehicle trajectories data of the roundabouts: (a) a quadcopter UAV (Figure 6) and (b) a RTK GNSS receiver. The selected UAV (DJI Phantom 4 Advanced) can capture videos up to C4K analysis with a frame rate of 60 fps and high-resolution images (5472 × 3078). It is a low-cost UAV and requires less experience to be controlled. Its camera is attached to a gimbal, which stabilize shots. Moreover, its weight (1368 g) and size contribute to better maneuverability, while low altitudes flights achieve high spatial resolution. The RTK GNSS receiver that was used provides reliable and high-accuracy data collection. The position of selected GCPs can be determined in centimeter-level accuracy in real world conditions. Specifically, the accuracy of this equipment examined for a survey is 8 mm + 1 ppm (horizontal) and 15 mm + 1 ppm (vertical).

**Figure 6.** (**a**) The quadcopter UAV and (**b**) the captured study area as presented on the controller during the survey.

#### *3.2. UAV Survey*

Roundabouts' performance was recorded with the use of the Unmanned Aerial Vehicle (UAV) during summer and autumn of 2019. Data regarding the kinematic characteristics of the vehicles were collected in this experiment.

Field measurements were selected to be conducted during of-peak periods to ensure free flow speed conditions. Real vehicle speeds in unobstructed traffic conditions were collected. Weather conditions were stable and did not affect the vehicle movements. Low flight altitudes of a nadir point of view and high-resolution recordings allowed to measure accurate and naturalistic spatiotemporal phenomena of high detail.

#### *3.3. Data Processing*

Vehicle trajectories were extracted from the UAV videos following specific steps. These steps aimed to: (a) reduce any bias and increase the accuracy and (b) digitize and study vehicle motion paths.

Firstly, no significant frames from videos (take-off and landing) were removed. A stabilization procedure was followed [40]. Proper filters made videos smoother while camera shakiness was eliminated. Specific events were identified (vehicle through movements on free flow speed conditions) and video frames were extracted for each examined video. Lens distortion of each frame was corrected [41]. The acquired images were georeferenced in high accuracy using an open-source GIS software [42]. The selected geographic reference was the "GGRS87/Greek Grid". The ground sample distance was measured to the value of 35.6 mm and the maximum value of the RMSE between the real location and the final georeferenced location per each frame was 0.6.

Finally, a digitization process was carried out to acquire vehicle trajectories. Specifically, the center of the front bumper of each vehicle was identified and the coordinates were extracted. Mathematical interpolation based on the known coordinates was used in CAD software using spline curves so the discrete data could be transformed into continuous functions to further study vehicle maneuvers and microscopic traffic data.

#### *3.4. Experiment Results*

Several applications regarding microscopic traffic data analysis can be carried out regarding the extracted dataset. Understanding the way road users move in roundabouts is of great importance, since many conflicts occur in this type of intersections [43,44].

Figure 7 presents a heatmap of the spatial distribution of vehicle maneuvers according to the extracted vehicle trajectories of the examined case study roundabout. Various driving behavior patterns for through movements on the multilane roundabout are indicated. Improper and unexpected paths are created and that can probably cause unexpected behavior, and therefore potential danger.

**Figure 7.** Heatmap of the spatial distribution of vehicle maneuvers of Roundabout A. **Figure 7.** Heatmap of the spatial distribution of vehicle maneuvers of Roundabout A.

traversing the roundabout is presented in Figure 8.

To calculate vehicle speeds at the entrance of the roundabout, firstly the proper events were identified. Then, the captured distances per 0.2 s were calculated and divided by the corresponding time. Considering the measured value of the ground sample distance (35.6 mm) and the RMS error of 0.6, it is concluded that the maximum error in speed estimation is less than 1 km/h. An indicative calculated speed profile of a vehicle traversing the roundabout is presented in Figure 8.

**Figure 8.** A typical vehicle speed profile has been extracted by processing UAV shots.

The vehicle speed profile indicates significant changes in speeds because of the complexity of the road geometry. The driver reduces the vehicle speed along the entry and the center of the roundabout and then accelerates to the exit lane.

With the second case study roundabout, an analysis regarding the travel time of 31 vehicles passing through the roundabout was conducted. An algorithm was developed and applied to the extracted trajectories. Two detection lines were coded at the entrance and the exit of the roundabout to identify the timestamps of each vehicle passing these sections (Figure 9). Results provide information regarding the travel time of vehicles through-movements. The average travel time of through movements under free-flow speed conditions is 5.14 s.

**Figure 9.** Graphical representation of the vehicle trajectories and the location of coded detectors. **Figure 9.** Graphical representation of the vehicle trajectories and the location of coded detectors.

#### **4. Conclusions**

Research into collecting and measuring reliable, accurate, and naturalistic microscopic traffic data is a fundamental aspect in road network planning scientific literature. According to the literature review and the bibliometric and visualization analysis, it can be concluded that UAVs are recently being used in the transportation field to monitor and analyze the traffic flow. The vehicles' detection and classification to extract trajectories for operational and safety analysis is the main issue the existing literature is dealing with.

The proposed methodology of this paper is structured in such a way to be simply and effectively adopted by researchers and engineers. Detailed and naturalistic vehicle trajectories data can be extracted through UAVs in a low-cost way, requiring less demand for high skills or expertise in image processing techniques.

Two experiments were carried out on different light conditions following the presented framework. Results indicate that the accuracy of the extracted microscopic traffic data is high enough as the maximum error in speed estimation is less than 1 km/h. The measured data can be interpreted and applied to several studies. In these experiments, vehicle trajectories were analyzed to extract vehicles speeds, travel time measurements, and the spatial distribution of vehicle maneuvers. The extracted dataset can be utilized and be applied on various traffic studies (gap-acceptance analysis, road safety analysis, study on lane-change behavior, calibration of car-following models, etc.).

The limitations of the study are related mainly to the technical issues of the UAV technology and the climate characteristics of the examined area. More specifically, the restraint of the low battery duration of UAVs and the weather conditions of the study area (rain) resulted in short video recordings and time-consuming field experiments.

#### **5. Discussion and Final Remarks**

Accurate and high detailed naturalistic microscopic traffic data are essential for reliable traffic analysis and efficient calibration process of traffic models. The forthcoming applications of intelligent transport systems on vehicles and infrastructure require sufficient tools to calibrate existing models on more complex situations.

UAVs are one of the most emerging technologies being used recently in transportation field to monitor and analyze the traffic flow. The increased commercial market share of this technology during the last years and the developed traffic applications entail a promising contribution to the transportation sector.

Several applications and methodologies regarding the acquirement of microscopic traffic data have been tested until now. This paper aims to add value to the existing literature. A comprehensive framework for extracting naturalistic vehicle trajectories recorded by UAVs is proposed. The presented methodology describes a highly accurate and low-cost method to conduct UAV traffic surveys and extract the required information, considering the methods and the limitations reported in previous studies. Researchers and engineers can efficiently apply the structured process as it is simplified and does not require great skills or expertise in automated image processing techniques.

Overall, the benefits of using UAVs on microscopic traffic data collection surveys, as explained in the literature review and highlighted in the presented experiment are great of importance. However, there are still many limitations and issues that need to overcome during the next years to optimize the general process of UAV traffic surveys. Thus, the ability to select and implement UAVs as an efficient traffic survey tool depends on many factors.

Further research is required regarding the acceptability of UAVs on traffic data collection surveys. Specifically, a comprehensive guiding framework under what circumstances this technology should be preferred or discouraged compared to other traditional methods will enhance the overall process of the examined traffic analysis. Finally, due to the forthcoming applications of intelligent transport systems on vehicles and infrastructure, experiments on the capabilities of UAVs regarding real-time traffic data collection techniques and methods of sharing the traffic information to road users are expected to enhance UAVs technology in the transportation engineering sector.

**Author Contributions:** A.A. and F.K., conceptualization; A.A., methodology; A.A., writing; F.K. review, supervision. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Informed Consent Statement:** Not applicable.

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