Design And Realization Of Infrared Image Anti-Occlusion Tracking Algorithm Based On CC-KCF

Target tracking has been widely used in intelligent monitoring,intelligent transportation,human-computer interaction and military field,especially for small targets tracking in complex infrared scenes.In the infrared scene,the target tracking becomes difficult because of the abnormal situation,such as the target is occluded,and is disturbed by a strong radiation source(including different wave bands).This paper analyzes some algorithms of different tracking algorithms.According to the characteristics of infrared target,we propose an infrared image anti-occlusion Tracking Algorithm Based on the Consistency of Classifier of Kernelized Correlation Filter(CC-KCF).We design and realize the prototype of the tracking algorithm and set the corresponding technical parameters.The improved CC-KCF algorithm is applied to the multi-core DSP platform,which makes the computation speed of the image tracker greatly improved.In the stage of sequence image preprocessing,the characteristics of infrared image are analyzed and a method of quantization of weighted linear filtering based on adaptive parameter change is proposed.This method improves the anti-infrared radiation interference ability of the image sequence,so that the illumination change between the sequence images tends to be stable,which greatly makes the image quality better.In the research stage of tracking algorithm,this paper proposes a tracking algorithm based on the consistency of classifier of kernelized correlation filter,CC-KCF algorithm.The method of target occlusion detection based on kernelized correlation measure is proposed to determine whether the target is occluded or not,which can be "alarmed" in time after the target is occluded and it will come in the stage of re-capture.For target re-capture,matching based on the consistency of classifier of kernelized correlation filter is used to retrieve the target,which makes the anti-occlusion performance greatly improved.A multi-scale filtering method is proposed to make the tracking algorithm accurate and timely response to the scale of the target.The feature normalization method is proposed to enhance the algorithm's resistance of light mutation.The improvement of the kernel tracking algorithm makes the system's long-term stable tracking performance greatly improved.Finally,the CC-KCF algorithm is compared with other tracking algorithms under the same conditions.The algorithm is evaluated from the aspects of anti-occlusion,anti-mutation of light and tracking stability.In the end,the CC-KCF algorithm is implemented on the DSP platform.