Detecting and estimating the levels of crowd density from UAV imagery

Abstract EN

In recent years, unmanned aerial vehicles (UAVs) have increasingly been used in many civilian applications such as environment monitoring, aerial crop surveys, road monitoring, counting wildlife and fire detection.

One of these applications is crowd monitoring and management especially in highly crowded areas such as sporting events and Hajj seasons.

In this paper, color-based approach has been employed for detecting and estimating crowd density using UAV images.

The Analysis of UAV images were performed under different cases of camera position and orientation including vertical, horizontal and close range camera.

The results show that the method was able to detect the crowd among other image objects (e.g., buildings, vegetation and streets).

In addition, the results were improved under vertical camera case comparing with those under close range or horizontal cases.

This is due to the fact that the camera was focused on the crowd area more than other surrounding environment.

Furthermore, the method performed well in Hajj images comparing with other images cases.

This is because the number of feature classes was limited (mostly black and white).

In order to improve the results, the associated anomalies/outliers have been eliminated from the segmented image through using image enhancement tools.