Real-time monocular image-based path detection

PABLO DE CRISTÓFORIS; MATÍAS NITSCHES; TOMÁS KRAJNÍK; MARTA MEJAIL

Journal of Real-Time Image Processing

Springer-Verlag

Lugar: Berlín; Año: 2013 p. 1 - 14

1861-8200

In this work we present a new autonomous navigation method. This method allows a mobile robot equipped with a monocular camera to travel through different naturally delimited paths. It does not require camera calibration or a priori given knowledge of the environment. The core of the method is based on segmenting images and classifying each super-pixel to infer a contour of navigable space. The contour shape is used to calculate the forward and steering speed of the mobile robot. To achieve real-time response of the robot, the image segmentation is implemented on a low-power embedded GPU. The validity of our approach has been verified with an image dataset of various outdoor paths as well as with a real mobile robot.