Construction of 2.5D Map of Dynamic Environment Based on Information Fusion

When using Lidar for map construction, it will be affected by the dynamic objects in the environment, reducing the accuracy of the mapping, and two-dimensional Lidar can only scan the structural information of a single plane. Using such a map will cause the robot to collide. The Kinect v2 is an RGB-D camera, which can collect depth information and color images in the environment, but its accuracy is low, and it is greatly affected by lighting during use. In order to reduce the impact of dynamic objects in the environment on mapping, we propose a method of using convolutional neural networks to eliminate dynamic objects. Then propose a method of kinectv2 and Lidar fusion to build a two-dimensional map. Experiments were conducted using the robot operating system ROS, and the experimental results show a very impressive improvement with respect to the traditional method.