Research On Binocular Image Matching Of Camellia Oleifera Fruit Based On Improved SIFT Algorithm

The application of the picking robot is helpful to alleviate the problem of the growing shortage of agricultural labor,and can greatly improve the efficiency of agricultural products picking.However,how to match images quickly and realtime in the process of picking is related to the application and development of picking robots in agricultural production.Therefore,it is of great practical significance and wide application prospects to study the algorithm for matching the binocular image of camellia oleifera.This paper covered the following aspects:1.The basic principle of image matching and the parallax theory and coordinate transformation of binocular imaging are studied,a simplified camera calibration method are discussed and analysed,and the calibration experiment of the camera are completed based on this.At the same time,the system of stereo matching elements is analyzed.The content of stereo matching is analyzed from four aspects:feature space,search space,search strategy,and similarity measurement.The matching algorithms are summarized from three different classifications:matching the degree of density,the scope of action and the matching base element.2.Because the working environment of Camellia fruit picking robot is in unstructured environment,and the quality of the camellia fruit image is difficult to guarantee.Therefore,the camellia fruit image pretreatment is studied,the enhancement method of the original camellia fruit image is analyzed.Contrast enhancement is used to realize the contrast enhancement of the original camellia oleifera image.And then the denoising of fuzzy camellia fruit images with dryness sound is studied,several commonly used denoising methods are used,and it uses an improved threshold-resolving wavelet denoising method for denoising.Experiments show that this method can get better noise reduction effect.At the same time,the edge extraction,binarization and image correction of the image denoising are also carried out.3.The SIFT matching algorithm is studied,the steps of the SIFT algorithm are given,and the various steps are introduced in detail.Based on this,we improve the real-time performance and propose an propose an improved SIFT feature matching algorithm based on adaptive threshold feature matching and building an image pyramid.4.The performance of the SIFT algorithm is verified in an actual unstructured environment.The results show that both the rotation invariance test,the noise invariance test,and the brightness change invariance test all achieve ideal results.At the same time,a validation experiment is conducted to improve the real-time performance of the SIFT algorithm.The match verification experiment of the collected camellia fruit images was carried out from four aspects:feature point extraction,time consumption,correct matching rate,algorithm processing speed and different lighting conditions.The improved algorithm presented in this paper is faster than the SIFT algorithm.Although the number of detected features is relatively small,the average time is only about half that of the SIFT algorithm,and the real-time performance of robot picking and positioning is more potential.At the same time,in the three different cases of scale transformation,rotation transformation and luminance change,the improved algorithm's correct matching rate,feature point detection time and matching time are all better than the original SIFT algorithm.Moreover,the improved SIFT algorithm proposed in this paper shows better matching effect than traditional SIFT algorithm under different lighting conditions.Under the appropriate light intensity,ie,1 to 12,000 1x,the matching rate meets the actual application requirements,and it has stronger unstructured environmental adaptability than the traditional SIFT algorithm.