【摘要】 监控视频被广泛地应用于刑事侦查中,对监控视频进行显著性检测,能够充分挖掘和分析监控视频中的关键刑侦信息。传统的显著性检测方法适用于自然图像和视频,目前并没有面向监控视频的显著性检测方法。监控视频显著性检测作为一个新的应用领域,有许多问题尚待解决,其中最为关键的是如何选取合适的度量突出具有较高刑侦价值的区域。针对上述问题,提出了一种新的基于刑侦关注模型的监控视频显著性检测方法。首先,计算刑侦关注对象的出现概率,作为刑侦关注度的度量,使用该度量计算得到空域显著度图。然后,引入时域的运动信息,融合得到最终的时空域显著度图。实验结果表明,提出的方法能够准确快速地确定监控视频的刑侦关注区域。更多还原

【Abstract】 Surveillance video is widely used in criminal investigation, detecting the salient regions in surveillance video can fully exploit and analyze critical information in the surveillance video. Traditional saliency detection method is applicable to natural images or video, and there is no saliency detection method for surveillance video. As a new application field, there are still a lot of problems to be solved in surveillance video saliency detection, and the key is how to choose the appropriate measurement to highlight the area with high value of criminal investigation. To solve this problem, a new saliency detection method based on the criminal investigation attention model is proposed. First,we calculated the occurrence probability of the criminal investigation attentive objects, as a measure of the degree of criminal investigation attention, and utilized it to obtain the spatial domain saliency map. The final spatial-temporal saliency map was obtained by introducing the motion information in the time domain. Experimental results demonstrate that the proposed method can accurately and quickly detect the criminal investigation attentive area of surveillance video.更多还原