For the public security, visual data processing is an important research direction of the Internet of Things. During the case investigation process, video surveillance data is an important basis and source of clues. Directly for the application of cases video judged, this thesis collects the video clues scattered in the non-overlapping monitoring network, use the tags to describe the characteristics of the suspects. Based on the tag data, research the trajectory of suspects, and assist police to determine the identity of the suspects. The specific contents are as follows: First, the video data of case is collected and tagged. The video data of case is collected from the non-overlapping monitoring network of crime scene and the nearby area. For the video, collaborative tagging system describes the characteristics of the suspects, centralized storage to clue database with the form of tag data. Then, the tag data is processed. Because of the tag data with high-dimensional mixed-type, in accordance with the attribute type, tag data is divided into three sub-space of continuous, boolean, enumeration type. Similarity of sub-space is calculated using three different methods, the combination of the subspace calculating the weighted average is the similarity matrix of suspects. With cohesion-hierarchical clustering algorithm, similar suspects are clustered into one group. Combined with time and space information, analyze the trajectory of the suspects. Finally, contrast the trajectory of suspects and the signal trajectory of the cell phone, determine the phone number to trace the real identity of the suspects. Discovering valuable clues from a massive video data help police judge information to solve the case. The application of collaborative tagging system is depth description of the case of video data, with the tag metadata to describe the suspects. Tag data mining shows the trajectory of the suspect. Not only has theoretical research value, and also it has clear practical value in the field of the public security.
修改评论