Daytime Water Hazard Detection Based On The Multi-feature Fusion In The Wild Field

This paper summarizes the author's, achievements during thepursuing of master's degree. The project which detects hazards in the wildfields is supported by National Natural Science Foundation and it is basedon multi-sensor fusion. We get the validity of the algorithm according tolots of experiments.Autonomous land vehicle (Autonomous Land Vehicle .listed ALV)is an intelligent autonomous mobile robot. It includes a lot of theories andtechniques which involves a number of disciplines. ALV reflects thelatest achievements of science and artificial intelligence technology.The capability of reliable hazards detection is the precondition ofintelligent mobile robot's safe navigation in the complex environment.Varity of the hazards and natural environment brings a great challenge.Ponds and other water bodies are high security threats to the intelligentrobot's navigation. In order to detect water, we use a method which is based onmulti-feature fusion in this paper. For the no reflection area of the waterin the wild environment, we use brightness and texture. Water hazard isgenerally the area of high brightness and weak texture. For the reflectionarea, we use stereo vision to detect water area. The height of thereflection is lower than the height of the ground in three-dimensioncoordinate system. We can use this feature to detect water area. Finallywe fuse the brightness, texture and reflection characteristics to detectwater hazard. The experimental results can segment the water hazards ina picture accurately and verify the effectiveness of this method.This paper also presents a machine learning (Support VectorMachine) method to improve the detection of the water hazards. We tryto use machine learning to detect brightness feature, and get satisfactoryresults.