Vision-based detection and distance estimation of micro unmanned aerial vehicles

---

Tezin Türü: Doktora

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Bilgisayar Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2015

Öğrenci: FATİH GÖKÇE

Danışman: GÖKTÜRK ÜÇOLUK

Özet:

In this thesis, we study visual detection and distance estimation of Micro Unmanned Aerial Vehicles (mUAVs), a crucial problem for (i) intrusion detection of mUAVs in protected environments, (ii) sense and avoid purposes on mUAVs or on other aerial vehicles and (iii) multi-mUAV control scenarios such as environmental monitoring, surveillance and exploration. The problem is challenging since (i) a real-time solution is required, a burden when computational power is limited by the hardware carried by an mUAV, (ii) non-convex structure of the mUAVs causes the bounding box of mUAVs to include very different background patterns, (iii) background patterns from indoor or outdoor are very complex with different characteristics and can include moving objects, (iv) mUAVs tilt and rotate unavoidably resulting in very large changes in their appearances, (v) when the camera is not stationary, motion blur is a problem, and (vi) illumination direction and brightness changes cause different images. We evaluate vision algorithms for this problem, since other sensing modalities limit the environment or the distance between the mUAVs. We test Haar-like features, Local Binary Patterns (LBP) and Histogram of Gradients (HOG) using boosted cascaded classifiers. We also integrate a distance estimation method utilizing geometric cues with Support Vector Regressors. We evaluated each method on indoor and outdoor videos collected systematically and on videos with motion blur. Our experiments show that, using boosted cascaded classifiers with LBP, near real-time detection and distance estimation of mUAVs are possible in about 60 ms indoors (1032x778 resolution) and 150 ms outdoors (1280x720 resolution) per frame, with a detection rate of 0.96 F-Score. However, classifiers of Haar-like features lead to better distance estimation since they position the bounding boxes on mUAVs more accurately. Our time analysis yields that classifiers of HOG train and run faster than the other algorithms.