Study On Image Collection And Recognition Technology And Application

Image collection and recognition is a significant part of computer vision. Image collection mainly includes selection of image sensor, adjustment of light intensity, aperture and exposure time, design of image collection platform and image preprocessing technology. First, suitable image sensor should be chosen to ensure the image collection quality. Otherwise, light intensity, aperture and exposure time should be adjusted. It is important especially for moving target. Next, an image collection platform should be built. It is for the fixing of light source and camera. Finally, the image must be preprocessed to highlight the interesting part and make it convenient to be recognized. Image recognition is to extract the characteristics of image, understand the information reflected by characteristics, and it is the basis of target classification.This paper studied on the effect of image collection environment on the image quality of moving target. Chose badminton wind tunnel test as research target, optimized the image collection environment, corrected the motion blur of image, extracted the characteristic of image and finally designed a badminton classifier based on SVM. The main work of this paper as followed.1. Research on the effect of image collection environment on image quality. Image collection environment will affect the image quality straightly. This paper compared CCD and CMOS image sensor, point out the pros and cons of both sides. Aim at the moving target, selected the image sensor. Analyzed the effect of light intensity, aperture and exposure time on image collection of moving target, and optimized the platform of badminton image collection.2. Research on image preprocessing of moving target. For image of moving target, blur is unavoidable. Because of the motion of badminton is rotational, Wiener filtering cannot be used to recover the blurred image straightly. At the condition of rotational speed is already known, traditional rotational motion blur recover method required the perpendicular of rotational axis and capture plan. This paper raised a rotational motion blur recover method based on estimating the angle between rotational axis and capture plan. 3. Image characteristic extraction The characteristics of image can be extracted after the preprocessing. The research target badminton’s characteristics are diameter, swing and rotational speed. This paper used the shape of badminton to measure its diameter, analyzed the image sequence to measure the swing, and based on the change of characteristic point which is on the ball head in entire image sequence to estimate the rotational speed of badminton.4. image recognition and classificationThe characteristics extracted form image sequence was analyzed in this paper. The structure of classifier based on decision tree was optimized. Based on SVM, the classifier was designed by using future knowledge to classify the grade of badminton. And the classification error was analyzed at the end of this paper.The image processing system in this paper was programmed based on Matlab and VC, realized the analyzing of badminton pose in wind tunnel and classification of badminton, and has some theory meaning and good use value.