Flight Tests of the Vision-based Target Sensing and Approaching

Abstract

In this paper, we propose the visual target sensing algorithm using the feature fusion and validate by flight tests. Sensing algorithm is divided by two parts: multi-feature detector and fusion. Multi-feature detector consists of the point-like feature, texture, and shape detector. The fusion part is implemented in the sequential update step of the Kalman filter-based tracking algorithm. This approach can detect the target robustly in the real outdoor environment that the scale of the target is changing. Furthermore, we designed the time-to-go-based longitudinal guidance controller with the direct visual servoing and performed the Open-loop and closed-loop flight tests by using the quadrotor and fixed-wing UAVs. The open-loop test is performed by using the quadrotor UAVs for validating the visual sensing algorithm; the closed-loop test is repeatedly performed by using the fixed-wing UAVs with the direct visual servoing-based guidance in the same environment. Finally, we analyze the results that approaching the target based on the circular error probability (CEP).