| In the area of public security,video monitoring has become a crucial protective barrier in many industry safety systems.However,a large number of video monitoring systems still rely on manpower.The application of target detection and tracking technology in video monitoring can help us save a lot of labor costs.However,this still faces many challenges,such as poor lighting environment,target scale changes and so on.Firstly,this paper describes the basic principles of DPM target detection algorithm and KCF target tracking algorithm,and analyzes their shortcomings in the application.In light of the lack of illumination,a fast low illumination enhancement algorithm is proposed,and its empirical parameters are optimized by particle swarm optimization(PSO)algorithm.It can adapt to monitoring video application scenarios.In the target detection section,this paper puts forward to extend gradient component in the process of HOG feature extraction.It fully considers the effect of diagonal region on the gradient amplitude and gradient direction of the center pixel.In addition,the average accuracy rate of the Deformable Part Model(DPM)target detection is improved by combining Gabor transform and HOG feature.A target detection and recognition strategy combining frame difference method and DPM is proposed,which reduces the detection range of DPM and improves the efficiency of algorithm.In the target tracking section,the method of using feature pyramid is proposed to improve the self-adaptability of KCF to target scale change.In addition,a reference condition is proposed to judge the deterioration of the target tracking effect.Through this way,it can judge whether the KCF target tracking results are credible.A target tracking system combined with KCF target tracking algorithm and Calman filter is proposed to improve the stability of target tracking.Finally,the whole process of target detection and tracking is expounded,and the above improvements are verified through the whole experiment. |
PDF Full Text Request