Research On Image Mosaic Based On Depth And Color Dual Information Feature Source

The acquisition and processing of plant information is widely used in plant growth monitoring,precision spray target detection,and precise positioning of agricultural robot.The wide view image formed by image mosaic plays an important role in crop management and agricultural spray vehicle navigation.The traditional image mosaic method takes a long time to mosaic and is affected by the uneven illumination.There may be some problems such as duplication,dislocation and missing.According to the depth and color information obtained by Kinect sensor,this paper proposes an image mosaic method based on depth and color dual-source information,and carries out simulation experiment with MATLAB.Experiments show that the image after mosaic is smooth,and method has high matching accuracy and high mosaic efficiency.The main research contents are as follows:(1)Aiming at the problems of image dislocation,missing,ghosting and inaccuracy of traditional mosaic methods,a double information feature source image mosaic method is proposed.Kinect sensor is used to achieve images from different perspectives.Firstly,the color image is processed by SIFT to extract feature points and match feature points.Then,the plant depth information is used to eliminate mismatches.According to the correct matching pairs,the RANSAC algorithm is used to find the best projection transformation matrix H.Finally,the best stitching line algorithm is used to fuse the image,so as to achieve accurate image mosaic.Experiments show that the average matching accuracy of the traditional image mosaic method is 88.1%,and the average matching accuracy of the feature points of the image mosaic method in this paper is 92.9%,which is 4.8 per cent higher than that of the traditional method.Considering that the edge of Kinect will be missing when it is disturbed by sunlight,outdoor experiments are carried out.The average matching accuracy of traditional image stitching method is 92.1%,and the average matching accuracy of feature points of this image stitching method in this paper is 99.1%,which is 7.0 per cent higher than that of traditional image stitching method.The validity of this method is further verified.(2)A fast image mosaic method based on K-means clustering is proposed.Firstly,Kinect sensor is used to obtain plant depth and color information,and K-means clustering and depth information are combined to obtain effective plant area.Then SURF algorithm is used to extract feature points to reduce image mosaic time.Then feature points are matched,error matching is screened and projection is searched.Finally,multi-resolution image fusion method is adopted to fuse multiple images.In indoor and outdoor experiments,the average stitching time of this method is 3.52 s(indoor)and 7.11 s(outdoor),which is 8.62 s(indoor)and 38.56 s(outdoor)shorter than that of dual information feature source image stitching,and the average matching accuracy is 96.0%.(3)In order to improve the efficiency of image mosaic and ensure the integrity of the image,this paper proposes a fast image mosaic method based on local feature adaptation.Firstly,the vehicle image testing system is designed,and the vehicle running platform is built,and the frame rate is adjusted adaptively according to different vehicle speeds.Firstly,according to the gray histogram of the image,the color image is divided into four regions.According to the characteristics of different regions,the corresponding key points extraction method is adaptively selected.Then,the feature points matching and mismatching pairs are filtered.Finally,multi-resolution image fusion algorithm is adopted to fuse multiple images.The experimental results show that the average matching accuracy of this method is 96.1% and that of K-means clustering algorithm is 95.2%.Therefore,the matching accuracy of this method is still high.The overall average time-consuming of local feature adaptive fast image mosaic method is 3.29 seconds,which is 39.58 seconds less than that of dual information feature source image mosaic method,and 2.73 seconds less than that of K-means clustering fast image mosaic method.