Aiming at problems of robot scene recognition in indoor environment

a senmantic mapping algorithm is proposed

autonomy

realtime and accuracy of scene classification strategy are focused on. Two-dimensional grid map is constructed by mapping depth information and autonomously plan recognition path of scene. Convolutional network is applied to set up scene categorization model

recognize semantic classes without specific training in realtime. By Bayesian framework fusing prior knowledge

modify error classification and accomplish semantic mapping.Experimental results show that robot can carry out global autonomous exploration and realtime judge scene category

and set up semantic mapping which meets need. At the same time

in real path planning

robot can improve navigation behavior according to semantic information

feasibility of the method is verified.