Estimation of Human Body Height Using Consumer-Level UAVs

Abstract

Consumer-level UAVs are often employed for surveillance, especially in urban areas. Within this context, human recognition via estimation of biometric traits, like body height, is of pivotal relevance. Previous studies confirmed that the pinhole model could be used for this purpose, but only if the accurate distance between the aerial camera and the target is known. Unfortunately, low positional accuracy of the drones and the difficulties of retrieving the coordinates of a moving target like a human may prevent reaching the required level of accuracy. This paper proposes a novel solution that may overcome this issue. It foresees calculating the relative altitude of the drone from the target by knowing only the ground distance between two points visible in the image. This relative altitude can be then used to calculate the target-to-camera distance without using the coordinates of the drone or the target. The procedure was verified with real data collected with a quadcopter, first considering a controlled environment with a wooden pole of known height and then a person in a more realistic scenario. The verification confirmed that a high level of accuracy can be reached, even with regular market drones.

Keywords

• remote surveillance
• target height
• consumer-level UAV
• pinhole model
• image distortion compensation
• uncertainty analysis