Research On Real-time RGB-T Fusion Tracking Algorithm Of Embedded Motion Platform Based On Significance And KCF

In recent years,visual target tracking has been widely used in robot technology,automatic vehicle,human-computer interface and video monitoring,so it has been widely concerned.Although many important breakthroughs have been made in target tracking,the target tracking at present still faces many challenges,especially in various complex environment conditions(such as low light,rainy days,smoke,etc.),the imaging quality of visible image is significantly affected.Thermal infrared imaging can capture the thermal radiation from the target,not sensitive to the changes of light,but also has a strong ability to penetrate smoke,and can effectively supplement the shortcomings of visible light imaging,which makes RGB-T target tracking better meet the requirements of embedded motion platform real-time tracking algorithm.This paper studies the imaging characteristics of visible light and thermal infrared,combs the research status of RGB-T fusion tracking at home and abroad,analyzes the algorithm requirements of real-time target tracking on embedded motion platform,takes the video with platform motion features in VOT2019-RGBT public data set as the data set,and takes the average frame rate and accuracy rate as the evaluation index.Based on the construction principle of color model,this paper proposes a KCF tracking method(YUV-KCF)based on YCb Cr color model and RGB-T image fusion.In this method,infrared image is used to replace the brightness component of visible image,and KCF is used to realize target tracking based on the fusion image.Experimental results show that this method can keep the high-speed features of KCF algorithm to the maximum,and improve the accuracy of the tracking algorithm.This paper also proposes a RGB-T target tracking algorithm(KCF-LC)based on the fusion of the significance of gray level co-occurrence matrix parameters and KCF for the situation of blurred horizon when the moving platform tracks moving targets.This algorithm is based on YUV-KCF,aiming at the scene characteristics of KCF tracking algorithm when there is false detection,it is improved by LC track algorithm with strong false detection and re-search ability.In this paper,the gray level co-occurrence matrix parameter is selected to represent the scene blur caused by platform motion,and the feature vectors of visible and infrared images are calculated respectively for fusion.Experimental results show that,compared with similar algorithms,this method can solve the problem of target loss caused by blurred view when tracking moving targets on the moving platform,and has higher operation speed,which can better meet the requirements of real-time RGB-T fusion tracking on the moving platform.In order to realize the real-time RGB-T fusion tracking project of embedded motion platform,this paper builds a hardware platform of FPGA+DSP.This paper analyzes and optimizes the performance of the target tracking algorithm implemented on DSP,such as target loss and algorithm timeout.It strictly parallels the traversal operation and effectively meets the requirements of engineering implementation.