Unmanned vehicle water detection and tracking method based on SVM and SURF

Abstract

The invention discloses an unmanned vehicle water detection and tracking method based on the SVM and SURF. The method comprises the following steps of extracting color and texture features of a water obstacle; establishing a positive and negative sample bank of a water target and the ground surface, constructing a support vector machine classifier, and updating the search window position and scale based on an SURF algorithm; and detecting and tracking the water obstacle in a real-time video using a support vector machine algorithm. As the four texture features of the water are extracted and the water obstacle S / V is extracted as the color feature at the same time, the texture and color features are combined to constitute the water range features so the invention has higher detection accuracy as compared with a detection method using a brightness feature alone. In the invention, as the SVM algorithm is used for training the color and texture features to obtain the classifier, a kernel function RBF is used for transforming a non-linear problem into a linear problem in the high-dimensional space for solution, so the water detection accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of pattern recognition and image processing, in particular to a field water body detection and tracking method for an unmanned vehicle, which is suitable for detecting and tracking water body obstacles when the unmanned vehicle is driving in an off-road environment. Background technique [0002] Unmanned ground vehicle (UGV) is a typical outdoor mobile intelligent robot, and obstacle detection is one of the research hotspots of intelligent mobile robot. Compared with structured urban roads and expressways, the driving environment and road conditions in the off-road environment are more complex, and there are even no obvious road boundaries and drivable area boundaries. It also puts strict requirements on the performance of the UGV autonomous driving system, especially the environmental perception module. Various types of water bodies, as one of the most common types of obstacles in the field environment, ha...

AI Technical Summary

Problems solved by technology

[0006] All of the above methods have a certain impact on the real-time performance of obstacle tracking to a certain extent. Among them, the method based on model matching is more complicated in the calculation and analysis process and the update process of the target model, and it is more complicated to obtain an accurate model of the target, so it cannot meet the requirements. High real-time requirements; when the method based on area matching deals with the situation where the shape of the target object changes greatly or has a large area of ​​occlusion, the tracking accuracy of this type of algorithm for the target object will decrease, and even cause the target to be tracked. Missing

Images