Outdoor Road Detection Based On Online Adaboost Ensemble Learning Algorithm

The Accurate perception of road area is one of the most important prerequisites for intelligent mobile robots to perform outdoor work.In general,the roads can be divided into structured roads and unstructured roads.With the clear edge of the information,artificial lane mark and other reasons,the algorithms used to detect structured roads has made good progress.However these algorithms cannot apply to the detection of unstructured roads,which is mainly because the unstructured roads have the following characteristics: blurred road borders,various road types,rough road surface and imaging conditions(Constantly changing lighting and weather conditions can make the same scene into a variety of different color and texture information)which make the detection of unstructured roads have greater difficulty and higher challenges.In view of the above problems,this paper presents an outdoor road detection method based on online learning.To complete the road detection tasks under complex scenes,the key is to design a learning algorithm that can adapt to changing roads.The algorithm proposed in this paper can adapt to the changing of the roads.When the road changes,the training samples of the classifier can be updated in time to ensure the adaptability of the classifier.Its core thought is to build an off-line road sample set fist,calculate the similarity of adjacent images in the process of road detection,determine whether to update the online sample pool in the positive samples according to the similarity calculation result,where the online sample pool is used to train the updated classifier,and the source of the updated samples is the set of offline road samples.In order to ensure the accuracy and effectiveness of the algorithm,Adaboost ensemble learning algorithm is used in this paper,and the similarity calculation method based on color histogram is used to calculate the similarity.In this paper,experimental validation has made on the DLUT image data set and the Sowerby public image data set,the experimental results and data show that the proposed algorithm has high detection accuracy and practicability.Considering that most mobile robots are equipped with laser sensors,in order to further improve the road detection accuracy and make the road test results guide the movement of mobile robots directly,so trying to calibrate visual data and laser data,that a point in the space is not only with the image features also the cloud features.The features road sample and non-road sample are richer,which is beneficial to road detection.Finally,the road test results are directly mapped to the three-dimensional laser data,thus realizing the perfect combination of outdoor road detection and navigation of mobile robot.