Obstacle Detection Based On Multi-sensor Data Fusion For Duct Cleaning Robot

Multi-sensor obstacle detection and context-aware is one of the key technologiesof the central air-conditioning duct cleaning robots. The detection and identificationof the support pillar and the pipeline conditions about the main obstacles of thecentral air-conditioning have to proceed first in order to achieve the duct cleaningrobot’s autonomous navigation and obstacle-avoidance. According to the obstaclecharacteristics for central air conditioning and pipeline distribution characteristics,this paper mainly focuses on the integrated obstacles detection which based on cameravision coupled with other ranging sensors. The main contents of this paper are listedas follows:1. A monocular vision obstacle detection method based on vision has beenintroduced in this paper. Firstly we have to preprocessing (grayscale, histogramequalization, median filtering) the obstacle image before the edge detection of thetreated image by the method of wavelet modulus maxima algorithm. The experimentsprove that the wavelet modulus maxima algorithm have a better capability of resistingdisturbance compared with Sobel operator and the edge extraction is more effective.Finally, the actual length and width of the obstacles can be calculated by the edgeinformation of the obstacle image combined with software computation though thecamera perspective model and inverse model.2. The information fusion method which based on adaptive weighted fusionalgorithm has been investigated. According to the analysis of obstacles rangingsensors (ultrasonic sensors and infrared sensors) of air duct cleaning robot, themulti-sensor fusion model based on adaptive weighted fusion algorithm have beenanalyzed. The experiment result has obtained more accurate obstacle distanceinformation.3. The simple obstacle avoidance system of information fusion of the basic visualsensors and ranging sensor has been studied in this paper. Firstly, the actual size ofthe obstacles can be obtained by the vision sensors, and the distance of thesurrounding obstacles can be calculated by the ranging sensor, so the basic obstacleavoidance strategy can be put forward, and then the multi-sensor information has tofused by use the neural network. Finally, the results show that this method can beadopted to achieve simple obstacle avoidance under the obstacle environment for the air duct cleaning robots.