摘要
针对宽泛条件下不同视域场景多摄像机多目标的匹配跟踪问题,提出了一种基于纯目标的强鲁棒自适应SIFT(Scale Invariant Feature Transform)特征匹配算法。该方法为每个从视频图像中提取出的纯目标设置一个CamShift(Continuously Adaptive Mean Shift)跟踪器,利用自适应尺度空间因子提取目标的细节特征,采用基于BBF(Best Bin First)的双向匹配策略去除误匹配点,当目标的关键点数量太少,无法满足计算三维二次函数精确关键点位置时,构造了自适应尺度Harris角点检测法增补新点。通过对户外车辆、人员等在不同场景下的连续跟踪实验表明,本算法实时性好、自适应能力强,与其他算法相比,匹配耗时少,跟踪精度高。
According to the continuous tracking of multi-camera to multi-objective under the broad condition for different scene, this paper proposes a matching algorithm based on characteristics of the pure goal of robust adaptive SIFT （Scale Invariant Feature Transform）. This method establishes a CamShift （Continuously Adaptive Mean Shift） tracking device for each pure goal which withdraws from the video image. It uses adaptive criterion space factor to get detail characteristic of goal. It uses the bilateral matching strategy based on BBF （Best Bin First） to elimination the error matching points. When the quantity of the goal key points is too few to satisfy the calculation the precise key point position of computation three dimensional quadratic function, it designs the adaptive criterion Harris vertex examination law to supplement the new spot. The experiment of continuous tracking outdoor vehicles in different settings indicates that this algorithm timeliness is good and its adaptive ability is strong. Compared with other algorithms, this algorithm consumes less match time, but is high in tracking precision.
出处
《计算机系统应用》
2011年第9期107-111,173,共6页
Computer Systems & Applications
基金
福建省自然科学基金(2006J0414)
