The Research On Detection And Tracking Methods For Ochotona Curzoniae Based On Video Sequence

Ochotona curzoniae is one of the main creature calamities in Qinghai-Tibet Plateau and its neighborhood area. It is necessary to investigate Ochotona curzoniae to obtain the number of Ochotona curzoniae, the damage degree, and the way of behavior for Ochotona curzoniae prevention. The automatic behavior analysis by video recording can provide a new information method for investigation and prevention for Ochotona curzoniae. In the automatic behavior analysis system of the Ochotona curzoniae based on video recording, it is a key problem to detect accurately and track Ochotona curzoniae steadily in natual habitat environment. Due to the complexity of the color similarity between the target and background, and the random movement, the detection and tracking of Ochotona curzoniae in natural habitat environment are difficult. Based on the previous research, the detection and tracking methods for Ochotona curzoniae in natural habitat environment based on a video recording are studied. The main works are as follows:Firstly, the common methods of object detection were introduced briefly, and merit and demerit of all kinds of object detection methods were analysed. Then the normal level set methods were presentated. According to the problem of local binary fitting model falling into the local minimums easily during the evolution process, an improved local binary fitting model was proposed. In allusion to the issue that the contour can not be detected accurately by frame difference, the contour could be precisely obtained by image segmentaion methods, covering the shortage of setting initial active contour curve by manual operation. The detection results of Ochotona curzoniae in video sequence show that the proposed methods can detect ochotona curzoniae accurately and successively.Secondly, the notaion, classification and classification characteristic of motion target tracking were indroduced, and the Mean-Shift tracking method was indroduced highlightly. The adaptbility of Mean-Shift method for Ochotona curzoniae tracking was analysed. In view of the problem that the color between object and background was similar in Ochotona curzoniae tracking in natural habitat environment, a new visual descriptor based on texture information was proposed to reflect the subtle differences between the Ochotona curzoniae and the background. The visual descriptor was combined with color information to characterize the object model, and the object model was embedded into the Mean-Shift tracking framework for Ochotona curzoniae tracking. The tracking results of Ochotona curzoniae in video sequence show that the proposed method for characterizing the object has strong difference ability for target and background. Ochotona curzoniae can be accurately positioned under scenario of color similar between object and background.Furthermore, the basic theory, the type and the tracking methods of abrupt motion were presentated. The Markov Chain Monte Carlo and Wang-Landau Monte Carlo methods were analysed emphasisly. Due to the problem of randomness and unpredictability of movement for Ochotona curzoniae tracking, an Ochotona curzoniae tracking method based on the guidance of motion information was proposed. At first, the motion information was extracted between the adjacent frames used the frame difference method, and the movement mode was judged by motion information, then appropriate sampling tracking strategy was taken to track Ochotona curzoniae. The tracking results of Ochotona curzoniae in abrupt motion video sequence show that the method we proposed can not only ensure tracking performance of abrupt motions, but also improve tracking performance of smooth motions. According to the problem of color similarity between Ochotona curzoniae and background, and the problem of uncertainty and randomness of Ochotona curzoniae motion, the proposed target representation was embedded into the motion induction tracking framework. The experimental results show that the method can solve the tracking issue effectively under the scenario of color similarity between object and background and complex motion ways.Finally, the concept and anlysis methods of time series were presentated.The feasibility of wavelet neural network to predict nonlinear time series was analysed. A three-layer wavelet neural network was proposed for short-term Ochotona curzoniae behavior prediction. The forecasted results clearly show that wavelet neural network has good prediction properties for Ochotona curzoniae behavior prediction.