Indoor Trailing Detection Algorithm Design Based On Image Processing

Trailing is a common behavior before the crime, criminals can be moving together with the target, or they just stand close to the target, hovering around. These acts are often so common that it's difficult to be aware of. Although relying too close is not necessarily a crime, but if the security personnel can be reminded in real-time when suspicious behavior occurs, many possible trouble can be avoided. There are also many trailing detection flags, which can provide a great help in late retrieval with the video. A trailing detection algorithm based on image processing is proposed in this paper, the basic idea is detecting and tracking people in the scene to predict if there is trailing behavior.In the detection step, we scan the image and extract the HOG feature of a specific area, then judge it by the SVM method. We choose HOG feature due to its better invariance to changes in illumination or scale. We improved the scanning and feature calculating step to make the algorithm faster.Tracing step works by associating detected sequences and adding missing objects using Mean Shift method. Two properties are proposed and added to the tracking object: the number of continued frames and motion path, which make the detection step richer. The number of continued frames is the representation of credibility of the object, which can also help to eliminate false objects. The motion path of the object makes the detection step more effective, it's also the basis of trailing determination.Experimental results show that the accuracy of detection and the coverage of tracking meet the standard of trailing detection. Trailing detection experiments have also shown good results.