Duong-Van Nguyen, Lars Kuhnert, and Klaus-Dieter Kuhnert


Vegetation detection, time-of-flight sensors, 2D3D sensor fusion, feature fusion, outdoor navigation


Operating outdoor is still a daunting challenge for an autonomous ground vehicle, especially with the presence of vegetation. Therefore, this paper addresses a solution for general vegetation detection which lets the vehicle fully exploit its outdoor mobility capability. A new vision system integrated from photonic mixer device (PMD) and CMOS camera is implemented to provide simultaneously near-infrared (NIR), colour, and depth images. Whereby, the reflectance of the modulated NIR given by the PMD sensor and the red channel of the CMOS sensor are used to calculate normalized difference vegetation index (NDVI). In addition, a relative distance estimation method referencing a perfect flat ground is described to obtain quickly 3D point cloud in the vehicle frame, thus, enables a 3D distribution analysis for spatial feature extraction. As vegetation usually appears in several typical colours and unstructured texture, the paper also derives a methodology for generating colour histogram models and assessing unstructured texture orientation to create visual features. Finally, NDVI, spatial features and visual features are gathered to form feature vectors, which are then used to train a robust vegetation classifier. In all real-world experiments we carried out, our approach yields a detection accuracy of over 90%, which outperforms conventional approaches.

Important Links:

Go Back