The Research Of Optical Flow Algorithm And Its Application In Nuclear Explosion Video Surveillance

Computer vision obtains information, by imitating the human visual system.Computer vision system get what human want through collecting images, processingimages, and displaying feature information. Optical flow is an important researchdirection in computer vision, mainly be used for target detection, precision-guided,shape recovery and so on. Optical flow is widely used in the military and everyday life.Now there are five methods to calculate optical flow. They are differential method,matching method, energy method, phase method and neural dynamics method. Butthese methods’ widespread shortcomings are large computation, long computation time,and it is difficult to meet real-time requirements, so they are not used by the real-timemonitoring system in general. Scholars had carried out improvements to the oldalgorithms or proposed new algorithms to improve the efficiency of computation of theoptical flow field. The principle of differential method is relatively simple, and itscomputation is relatively easy, so there are a dozen of the improved algorithms based onit. This thesis studied several optical flow algorithms basing on differential method, andthen proposed that the optical flow field can be used in a nuclear explosion surveillance.Firstly, this thesis studies the Horn-Schunck algorithm, and then improves theiterative part of4-neighborhoods which is usually be used. Compares with theconventional algorithm, the improved algorithm reduces the number of iterations,improves the calculation speed. In addition, improved algorithm is added to differentialmethod that includes the image edge features which Xia Yupeng proposed. The resultsshows that this method reduces the number of iterations on the basis of convergenceedge.Then, this thesis studies the computation of feature optical flow, compares Harriscorner detector with sift feature point detection method, and uses Harris corner detectorto detect feature points of the image and match these points. Then I calculate the opticalflow field of the feature points with the improved Horn-Schunck algorithm. Thismethod has a small amount of computation, the calculation speed is relatively fast, andit gets a sparse optical flow field. Finally, this thesis calculates the optical flow field of nuclear explosion imageswith the improved differential method, analyzes the characteristics of the optical flowfield, such as the shape of the mushroom cloud, symmetrical horizontal speed, upwardvertical speed. According to these characteristics, we can determine that these acquiredimages are nuclear explosion images. This is another application of optical flow intarget detection and identification, and it has great practical significance.