Color Invariant Descriptors For Vision System In Forestry Robot

Forestry robot,belonging to the special industrial robot,can replace workers’ work in the complex forest work environment and greatly improve the work efficiency.Color information is a most intuitive and effective cue of objects,and also one of the important features obtained by the forestry robot vision system.However,color information is not stable.It is vulnerable to illumination variations,which may cause the robot vision system to obtain distorted color information,and then affect the performance of recognition and positioning.Our purpose is to make the forestry robot vision system have similar color constancy characteristics compared with human vision system,that is,the vision system can still recognize the real color characteristics of object with illumination variations.There are two methods to realize the color constancy of machine vision system: one is to estimate the illumination in the scenes,and carry on the color restoration processing to the image,that is,to calculate the color constancy;the other is to extract the color features with the illumination invariance in images,that is,the color descriptor.Based on the vision system of forestry robot,this paper focuses on the algorithm of color descriptor,which is invariant to light changes.The research is carried out from the following aspects:(1)Implementation of various color descriptorsIn this paper,based on the existing research status at home and abroad,five kinds of 30 different color descriptors are implemented,In this paper,five color datasets of indoor and outdoor scenes are selected for experiments.Indoor scenes are classified based on the surface characteristics of objects,and the relationship between different color descriptors and different surface characteristics of objects is studied and analyzed.(2)Establishment of a new color space coordinate system with illumination invarianceBased on Lambert reflection theory,this paper chooses the diagonal-offset model,which has good effect on indoor and outdoor scenes as the theoretical basis,to propose two new color space coordinate systems with illumination invariance: color information color space,edge information color space,and we proves its robustness to illumination variations from theory and experiment.(3)Derivation of a new color descriptor based on moment invariantsOn the basis of Hu moment theory,the color information component is introduced,and the color descriptors with invariance to illumination are derived in the new illumination invariance color coordinate system.The proposed color descriptors are applied to the field of image recognition and retrieval.The robustness to illumination changes is proved by a large number of experiments.(4)Comprehensive experimental evaluationThe color descriptors are evaluated by comprehensive experiments,which are divided into five color datasets for indoor and outdoor scenes.Indoor scenes are classified based on the surface characteristics of objects,which are divided into four categories: insulator surface,metal surface,fluorescent surface and surface with specular reflection.The relationship between different color descriptors and different surface characteristics of objects is studied and analyzed,and we compared the performance of different color descriptors for both indoor and outdoor image data in the context of image recognition and image retrieval.