Research Of Optical Flow From Color Image Sequences Based On Color Gradient

The objective of research on computer vision is to enable computer to take the place of human eyes and brain while perceiving, explaining and understanding the outside world. It can offer computer the ability to make use of one or even more images to recognize the environmental information around. Above all, the calculation of optical flow is the core in the researching field of computer vision and image-processing, which not only contains the moving information about objects being researched, but also brings itself with rich content about3D structure.It has been more than30years since the concept of optical flow was introduced. During its developing process, some effective computing methods had been created. Here are the five common categories:methods based on regional matching, energy, phrase, variational theory and wavelet. In most cases, these methods are being worked on the basis of gray information in the image, ignoring the large amount of color information.This paper first retrospects some classical gray and color computing methods of optical flow, the analysis of the principle of variational theory and relevant mathematics knowledge. Based on the principle of variational optical flow algorithm, the color gradient takes place of traditional gray to gain a calculation method of variational optical of color image sequence. Data terms adopt the combination of gray balance and gradient conservation, and absorbs local optimization named Lucas, so the final accuracy of illumination change and complex movement much improve. Smoothing term is the upgraded Horn smoothing,with smooth factors adapting the color gradient. It is used to regulate the balance between data terms and smooth terms,which greatly improves the property of calculation method of variational optical flow of color image sequence.In conclusion, this paper has proved the efficiency and accuracy of the computing method involved in the content through testing different groups and categories of color image sequence. By qualitatively analyzing the optical flow vectors and optical flow color information distribution curve,as well as errors of the calculating results contrast with other methods for quantitative analysis, the paper not only proves the validity and accuracy of the method, but also shows its superior property over other optical flow computing methods.