Detection And Tracking Algorithm Research On Infrared Image

Image tracking technology based on infrared sensor is widely used in our life,and how to achieve real-time and stable tracking has always been a hot issue.In this paper,the problem of infrared image tracking mesoscale change and target occlusion is studied.The main work includes the following sections:Firstly,the minimum output sum of squared error filter algorithm is researched and analyzed,and the HOG feature is used replace the grayscale feature for expanding the multidimensional direction.Secondly,in the scale change aspect,the target tracking is divided into two parts.The first part is responsible for determining the target in the entire image position,and the second part is responsible for establishing the scale gold tower search space in the determined position,and determine its optimal scale.Experiments prove that the improvement of this paper effectively solves the problem of scale change of targets.Thirdly,for long time tracking and target occlusion,this paper improves the MOSSE algorithm combined with the TLD algorithm framework.The target detection module and tracking failure judgement mechanism are added.When the tracking failure is judged,the target detection module is started and the target is initialized again.The detection module in this paper is a cascade detection method based on the combination of YOLOv2 algorithm and MOSSE algorithm.Finally,we conduct experiments and analysis the result.The first part is the evaluation system analysis of target tracking algorithm points,including data sets and evaluation indicators.The second part is the tracking effect comparison between our algorithm and the MOSSE algorithm.The third part is the comparison between the algorithm and other excellent algorithms.The results show that the tracking accuracy and robustness of the algorithm are better than other algorithms.