Research On Illumination Estimation Algorithms For Forestry Robots

Forestry robot belongs to the category of special robots,which can replace artificial activities and plays a very important role in forestry operations.As the most intuitive and significant feature of objects,color has become an important feature for forest robots to recognize different objects.However,the color of the subject surface is easily affected by the scene illumination,which makes the image color obtained by the forestry robot distorted.In the field of machine vision,the operation of reducing or ideally eliminating the influence of light on the image so as to restore the color information of the image,which is called color constancy.For the special environment of forestry robots,how to get better color recovery is the research purpose of this paper.There are two ways to achieve color constancy.The first way is to estimate the RGB values of the scene illumination when the image is taken,then restore the color of the image according to the diagonal model,named illumination estimation algorithm;the second way is to extract the robust features of the image,named the color invariant.This paper focuses on the research of illumination estimation algorithm for forestry robot.In view of the scene where the forestry robots work,the following aspects are studied on the basis of in-depth review and analysis of illumination estimation algorithms:(1)An improvement approach for pixel-based illumination estimation algorithms is proposed,using L0-norm smoothing to increase the credibility of image color.(2)An illumination estimation algorithm based on image feature is proposed.This algorithm first extracts the features of each image in the image set,and then determines K images with similar characteristics to the test image according to the features of each image.Finally,according to the standard illumination RGB value of K selected images,the weighted average is calculated to obtain the illumination RGB value of the test image.(3)An illumination estimation algorithm based on scene classification is proposed.This algorithm first uses SVM to classify the scene type of the test image,and then uses the corresponding optimal illumination estimation algorithm to estimate the RGB values of illumination.Finally,in the image set which contains the forestry scene,the author puts forward three proposed algorithms with the classic illumination estimation algorithm,the results show that the effects of proposed algorithms are better than other algorithms',which means that the proposed algorithms are more suitable for forestry robots working environment.Among the algorithms proposed,the effect of illumination estimation algorithm based on image scene features is the best,followed by the illumination estimation algorithm based on scene classification,and finally the illumination estimation optimization algorithm based on L0-norm smoothing.