A Corner Feature Detection Algorithm for 3D Grid Map

Nowadays, multi-robots begin to gain advantages when performing tasks in more and more complex environments. However, how to merge maps without any priori information and mutual observation is very important and is still an unsolved problem. Extracting corner feature points from the map and complete the following up data association can help merge maps for multi-robots. This paper proposed an algorithm based on sliding window and 3D convolution operator to extract corner features. We established a sliding window and used the 3D convolution operator to screening the corner feature from a 3D grid map. The proposed algorithm was directly based on 3D grid maps making it can extract comer feature for subsequent map fusion. Experimental results in the Google public datasets and our own datasets validated that the proposed algorithm can extract the corner features from 3D maps precisely and effectively in a variety of environments.