Adaptive structured and unstructured road detection using LIDAR and dome-camera

---

Tezin Türü: Yüksek Lisans

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

Tezin Onay Tarihi: 2017

Öğrenci: SİNAN ÖZGÜN DEMİR

Eş Danışman: AHMET BUĞRA KOKU, ERHAN İLHAN KONUKSEVEN

Özet:

Robotic ground vehicles are widely used in different road conditions to perform various tasks under semi- or fully-autonomous operation. To accomplish the given tasks, the vehicle should detect road regions accurately. Also, for a successful operation, the mobile robot requires a quick adaptation for changing road conditions. The objective of this study was developing an adaptive road detection algorithm for a semi-autonomous mobile platform (GOAT). For that purpose three different methods were developed for a robust classification of road regions ahead of the vehicle in both constructed and unconstructed environments. In the first method, LIDAR sensor was used to detect road regions by utilizing the data with adaptive parameter sets, which were estimated by utilizing discriminative learning approach. The experiments in structured environment showed that accuracy (ACC) of the output increased, while the false positive rate (FPR) decreased compared to the constant parameter approach. However, for the tests conducted in unstructured environment desired results were not obtained. Therefore, the second road detection algorithm based on visual and range measurement data needed to be developed. By this algorithm, approximately 50% decrease in the FPR values for both structured and unstructured road conditions was observed by filtering the segmented point cloud based on the hue color channel values. In the third road detection method, an online supervised learning algorithm was developed, which used the outputs of the second road detection algorithm to create and/or update visual road models. In the conducted experiments, it was shown that road regions and general road boundary behaviors can be detected both in front and back directions of the vehicle independent from the road shape.