Multiple Ground Moving Target Detection Under Complex Background

Computer vision based multiple moving target detection is a cross-cutting research topic in many disciplines such as image processing and pattern recognition.Aiming at dealing with the problem of background complexity and target diversity in ground target detection tasks,a multiple moving target detection method based on cascade processing of spatial and temporal information is proposed.Firstly,a candidate motion region extraction algorithm based on forward-backward motion history image(FBMHI)is proposed.The algorithm can be divided into four modules,which are image preprocessing module,global motion estimation module,FBMHI calculation module and candidate motion region generalization module.FBMHI contains the motion information of the front and back frames,which not only ensures high recall,high precision and high location accuracy of candidate motion regions,but also improves the adaptability to background rotation,complicated background environment,target entry and exit,and partial occlusion.The experimental results show that the proposed algorithm has high precision and high target localization accuracy.Then,a multiple moving target detection algorithm based on spatio-temporal local information is proposed.By analyzing temporal information,FBMHI is utilized to extract candidate motion regions.On this basis,exploiting the difference between target ant the surrounding environment,the maximum color contrast(MCC)in the LAB space is proposed to calculate the objectness of the candidate motion regions,which are then screened by thresholding to obtain object proposals.After that,the feature of object proposals is extracted based on bag-of-words model,and classified by a multiclass linear support vector machine(SVM).The object proposals,which are classified as instances of predefined classes,are determined as the final detection results.The experimental results show that the proposed algorithm can precisely end effectively detect multiple moving target under complex background.Lastly,a multiple moving target detection algorithm based on spatio-temporal context is proposed.By Temporal context is utilized to extract candidate motion regions via FBMHI.Spatial context and target appearance information are exploited to calculate object confidence map via sparse coding based conditional random field(CRF)model.Targets are distinguished from background by object confidence of the candidate motion regions.In the sparse coding based CRF model,the use of sparse coding improved the feature representation ability of target local appearance,and the CRF model exploits local context information.In addition,in the off-line training phase,the joint training of CRF model parameters and the dictionary improves the discriminability of the dictionary.In the on-line detection phase,the algorithm is accelerated by block processing and parallel computing.The experimental results show that the algorithm has high detection accuracy in the case of incomplete object,such as target entry and exit,partial occlusion,and so on.