Research On Flame Detection Application Based On Saliency Detection

With the popularity of smart cameras and the rapid development of multimedia technology,how to quickly and accurately remove redundancy from massive images,it is particularly important to find the information that humans need and interest.The saliency detection of images is aimed at analyzing the image information to extract the most interesting areas of the human eye.It is not only used as a preprocessing process in many fields,but also widely used in various applications,so the research of visual saliency detection has far-reaching significance.The main work contents and research innovations of this paper are as follows:(1)The current saliency detection approach usually can't highlight the salient object completely and suppress the background effectively,so an algorithm combining objectness prior with absorbing Markov chain is proposed.Firstly,the boundary absorption nodes are extracted by the background prior and boundary connectivity,and the absorbing Markov chain is used to compute the preliminary saliency map.Secondly,based on the prior knowledge of the object,the position of the semantic object is estimated,and the results are merged into adaptive segmentation.The refined background seeds are computed by the integration maps as absorbing nodes for random walk to obtain the final saliency map.The experimental results show that the proposed algorithm can better highlight the salient object and suppress complex background.(2)Aiming at the problem that the Markov chain random walk will get lost and can't distinguish the boundary background and the central background,a saliency object detection model combining the update mechanism and the Markov absorption probability is proposed.Firstly,the upper and right boundary superpixel background nodes are extracted according to the background prior knowledge,and the Markov chain absorption probability matrix is used to calculate the saliency,and then updated by the update mechanism.Secondly,the absorption probability is obtained and updated for the node with the least similarity to the background.Finally,the final saliency map is obtained by filtering optimization.The experimental results fully demonstrate the effectiveness of the proposed algorithm.(3)Focusing on the reality,in order to accurately locate the fire source point and realize the fire warning,the saliency detection algorithm is applied to the flame detection of the surveillance video,and a real-time monitoring fire warning method based on the human visual attention mechanism is proposed,which will be static saliency.The detection algorithm is combined with the dynamic frame difference method.The experimental results show that the proposed algorithm describes the multi-scale spatial feature information through the saliency analysis method,which is more robust and can accurately identify and locate the flame to prevent fire.