This demo process the segmentation results from the algorithm of semantic segmentation and extracts the centroid of each class in the image so that the depth value of these centroids can be calculated in a targeted manner to achieve simple obstacle avoidance.