Measurement of classified points using stereo vision and techniques of segmentation in disparity map for detection of obstacles.

Resumo

This work presents an obstacle detection system using a stereo camera as a sensor and segmentation
techniques in occupancy maps identifying navigable and non-navigable regions. The disparity images are used to
build a 3D point cloud whose distances are calculated from the ground plane. Then, points that are in a range of
distances are classified as potential obstacles and recorded on an occupancy map. Clustering techniques are then
applied to identify obstacles to identify the position, size and speed of obstacles. To validate the algorithm
developed in this work, some real tests were carried out with a full-scale autonomous vehicle.