Research On Target Image Fast Mosaic For Spray Robot

With the development of agricultural informatization and intelligence,image mosaic technology plays an important role in farmland crop monitoring and spray robot operations.The traditional image mosaic method has some problems,such as slow mosaic speed,low accuracy of feature point matching,and ghosting.In order to solve these problems,this paper uses the color and depth dual source information of the plant to study the accurate and fast mosaic method of the spray robot’s target image.The research content is summarized as follows:(1)The experimental platform for spray robot target image mosaic is designed.According to the mosaic requirements of the target image,RGB-D sensor,a micro embedded computer and other components are configured on the basis of the whole structure of the sprayer.The color and depth dual source information of the target plant are obtained through the RGB-D sensor,and the micro-embedded computer performs dual source data processing and information interaction.At the same time,the software interface of the image mosaic system is designed on the Windows platform based on Visual Studio,which provides a software and hardware platform for the research of spray robot target image mosaic algorithm.(2)In order to solve the problems of ghosting and low registration accuracy in traditional SIFT image mosaic methods,an improved SIFT dual source image mosaic algorithm based on local features is studied.The color image and depth data are acquired at the same time through the Kinect V2 sensor,and feature points of different types of color image sub-regions are detected based on local features.The depth information is used to eliminate the mismatched pairs to obtain the best homography matrix.The stitching algorithm combined with the dynamic programming is used to get the fusion image.Experimental results show that this algorithm effectively solves the problem of ghosting,and the average accuracy of feature point matching is 90.15%,which is 1.24% higher than the traditional SIFT image mosaic method.(3)Aiming at the problem of low mosaic efficiency,an improved GMS-RANSAC dual source image quick mosaic algorithm is studied.Real Sense D435 sensors are used to collect the color and depth information of target plants.ORB algorithm is chosen for feature point detection and extraction.GMS algorithm,combined with image depth and distance invariance and two-way strategy,is used to eliminate mismatches.According to GMS neighborhood grid,the RANSAC algorithm is improved,and finally a multiband image fusion algorithm is used to achieve the accurate mosaic.The experimental results show that the average accuracy of feature point matching of this algorithm is94.79%,and the average mosaic time is 1.72 s.Compared with the improved SIFT dual source image mosaic algorithm based on local features,the mosaic speed is faster.