视频监控中多视角目标智能定位追踪方法研究

发布时间：2018-07-27 18:46

【摘要】：在视频监控中对目标进行定位追踪时,容易出现遮挡情况。当前定位追踪方法需预先检测出遮挡状况,再对目标位置进行校正,计算量大,且追踪结果不可靠。为此,提出一种新的视频监控中多视角目标智能定位追踪方法,针对每个摄像头采集到的图像,对其进行二值化处理,通过稀疏表示方法对目标进行定位。构建了视频监控中多视角目标跟踪的概率框架,针对每个智能摄像机,建立视频监控中多视角目标追踪模型。通过不同视角中的视频数据求出本地似然函数。为了避免维度灾难问题,对联合后验概率进行分解获取边缘概率乘积的形式,通过边缘概率求出被追踪目标在摄像机视角下的位置。依据消息传递机制实现信息融合,达到目标追踪的目的。实验结果表明,所提方法有很高的定位和追踪精度。
[Abstract]:When the target is located and tracked in video surveillance, it is easy to appear occlusion. The current location tracking method needs to detect the occlusion state in advance and then correct the target position. The calculation is large and the tracking results are unreliable. For this reason, a new intelligent location and tracking method for multi-view target in video surveillance is proposed. The image collected by each camera is binarized, and the target is located by sparse representation. The probability framework of multi-view target tracking in video surveillance is constructed, and a multi-view target tracking model is established for each intelligent camera. The local likelihood function is obtained from the video data from different perspectives. In order to avoid the problem of dimensionality disaster, the joint posterior probability is decomposed to obtain the product of edge probability, and the position of the target under camera view is obtained by edge probability. The information fusion is realized according to the message passing mechanism, and the goal of target tracking is achieved. The experimental results show that the proposed method has high accuracy of location and tracking.
【作者单位】：商丘工学院信息与电子工程学院;
【基金】：河南省教育厅科学技术研究重点项目(12B520047)资助
【分类号】：TN948.6;TP391.41

【相似文献】