Abstract:  Stereo depth cameras, RGB-D cameras and other environment construction sensors based on color images will produce large errors or even fail to measure in a complex indoor fire environment with a lot of smoke and darkness, leading to a serious decline of autonomous fire-fighting robots in the environmental perception and reconstruction capabilities, then it won't be able to effectively detect and bypass unknown obstacles. Besides, multi-line lidar is expensive and has low indoor utilization. In order to improve the environment perception and reconstruction capabilities of autonomous fire-fighting robots in indoor fire scenes, this paper proposes a method for indoor 2D environment reconstruction based on structured light depth camera. Firstly, using the IMU attitude sensor correct the depth data; secondly, use the Area-Threshold-Height-Filtering algorithm to distinguish obstacles and roads in the world coordinate system, then we can get the map of 2D point cloud. Finally, build the dynamic model of the camera and the robot, so that we can fix installation errors and update the global environment map. We tested the improved algorithm on the indoor autonomous fire-fighting robot platform. The results show that the improved algorithm effectively improves the environment reconstruction and perception capabilities of autonomous fire-fighting robots in the complex fire scene, and the robot can accurately avoid obstacles and quickly go to the fire area to put out the fire. The article has a great significance of autonomous fire-fighting robots about navigation and obstacle avoidance.

Key words: structured light depth camera, fire fighting robot, IMU, obstacle detection, reconstruction of indoor environment