Research On Road Image Segmentation Algorithm Based On Ensemble Learning Framework

The main purpose of road detection is to understand and analyze the image information from different traffic road scenes acquired by the sensor,it is an important means to ensure that the vehicle can drive on the travelable area.It is the cornerstone to realize that the intelligent traffic bring the convenience to people.And the road image segmentation is the primary link of road detection.Image segmentation extracts the meaningful and interested regions of the image by sharing a certain similar properties,they will generally show a significant difference in the aspect of characteristic between the regions and others,providing valuable information for the subsequent segmentation work.In real life,the traffic roads have their own complexities such like the lane identification lines,road shapes,obstacles and so on,which make the traditional road segmentation methods mostly limit at the gray image,and there are many problems like boundary loss,fuzzy and other quality problems.Therefore,with the machine learning method to understand the road traffic information,to achieve road image segmentation is a good choice.This paper attempts to use the super pixel algorithm in the natural images,exerting its ability to preserve precise boundaries.On this basis,we also study the problem of feature selection and classification,and improve the segmentation effect of the road image.In this paper,the following two aspects of the study are carried out:(1)Aiming at the traditional pixel-based image segmentation methods,a regional growth algorithm based on super pixel is proposed.The segmentation method based on super pixel is a learning direction with more natural and semantic feature.For solving the inaccurate problem of the segmentation based on pixel method,this paper select the approximation degree among the calculated pixel values in CIELAB color space,using the SLIC super pixel algorithm to generate the super pixel blocks with the equal size and the controllable number,obtain the edge information map of super pixel blocks.And then we use the marginal pixels for region growing of multiple seed points,standardize the growth rules,to overcome the defects of unable getting closed boundary.Ultimately,we can get one or more texture target area of the road image segmentation.(2)This paper proposes an image segmentation algorithm based on super pixel and random forest,which integrates the idea of the shallow machine learning model based on clustering and classification.The image is clustered by using the SLIC super pixel generation method to obtain the local feature sub-region of the image,and the characteristic expression of the homogeneous super pixel is improved.Then use the integrated classifier random forest method which is superior to the single classifier to classify,optimize the training sample selection parameters,and enhance the degree of fit and integrity of the border,so that it can improve the segmentation accuracy better for the texture areas which is smooth or more intense.