Research On Tracking Methods Of Dairy Goats In Sheep Farm Environment

Object tracking technology is a comprehensive technical discipline that integrates image processing,pattern recognition and machine learning.It has broad application prospects in agriculture,such as the detection and control of crop pests and diseases,monitoring and management of animal husbandry,and vision navigation agriculture robots etc.With the rapid development of animal husbandry,how to accurately track dairy goats and other animals in real time to strengthen their monitoring and management has become one of the important research tasks in the future.Based on the classic correlation filter tracking algorithm and particle filter tracking algorithm,this paper studies the features expression,anti-occlusion processing,model updating,and feature fusion,and tests the improved method with the collected diary goat videos.And the main content as follows:1)A dairy goat tracking method based on context information and superpixel level features was proposed.Under the framework of correlation filters and tracking principles,based on the kernelized correlation filter tracking algorithm,context information was added around the goats to establish a context model.When the dairy goat target was severely obscured and the appearance changed dramatically,the information was lost,and the context information was used to predict its position.Suppresses target drift due to false appearance information.Then calculate the response value of the image area and the background area of dairy goats to detect whether the appearance of the target changes drastically,and then determine the update frequency of the context model to minimize the problem of model drift.In terms of feature extraction,superpixel-level mid-level visual features are used in place of traditional low-level features such as gray values or textures to make the appearance of dairy goat targets more stable.2)A self-adaptive weighted feature fusion particle filter tracking method is proposed.The color histogram feature and the histogram of the gradient direction?HOG?of the dairy goat were extracted to represent the appearance model of the dairy goat.The complementarity between the two features was used to achieve accurate tracking of the dairy goat.At the same time,it is difficult for most fusion algorithms to use a fixed fusion weight coefficient to adaptto the changes of the surrounding environment in real time.This algorithm uses the“reliability”of each feature to adaptively update the fusion weight coefficient,and the similarity between the color histogram feature and the HOG feature is achieved.The metrics dynamically adjust the weight coefficients of the two features to better cope with changes in the target appearance of dairy goats,compensate for the invariability of feature weights,achieve effective integration of the two features,and improve tracking stability.The experimental results show that the Proposed₁ algorithm improves the accuracy and success rate by 3.6% and 9.3%,respectively,compared with the kernelized correlation filter tracking algorithm.The Proposed₂ algorithm improves the accuracy and the success rate compared with the particle filter algorithm using only the HOG feature.Compared with16.9% and 10.5%,the accuracy and success rate were improved by 19.2% and 16.8%,respectively,compared to the particle filter algorithm using only color features.At the same time,the proposed algorithm is faster and meets the requirements of real-time performance.It has certain application value for future animal monitoring and management.