Moving Object Detection In Moving Camera And Trajectory Analysis

Moving object detection is a focus of the intelligent video surveillance system these days, and is a precursor to higher-level objectives such as object recognition, tracking and behavior analysis. For video surveillance tasks, the imager is often stationary and therefore automatic detection of moving objects is relatively easy since the model of the background is known. Background model estimation and subtraction techniques have been widely used in such cases. The moving object detection task becomes more challenging when the camera (observer) moves as well, due to the background"motion"which is induced by the camera. Cameras in the systems inevitably capture video full of irregular moving background, such as Robot navigation, auto driving technology, Unmanned Aerial Vehicles. Traditional background estimation and subtraction techniques do not apply in such circumstances.Against this background, the paper makes a research into object detection algorithm from moving camera. We propose a novel method of Moving Objects detection based on SIFT features matching and dynamic background modeling from moving Camera. It is found that SIFT is invariant to image translation, scaling, and rotation, and partially invariant to illumination changes and affine projection, which is benefit to image matching. We can take advantage of these good points to find the accurate parameters of the affine model, which solve the problem of the movement of the background caused by moving camera. Then, we adopt background subtraction based on dynamic update background model to perfectly detect the foreground object with shadow and ghost detection and removal. The paper makes use of the MHI method to make further research on the moving objects, gains some attributes of movement about the moving objects. We can understand the state of the movement deeper.The paper also does a great amount of work on the trajectory analysis. We choose the DTW distance as the similarity measure for the variant length of trajectory to detect abnormal trajectory of people and vehicle. We can more clearly know the state of the trajectory by taking direction and rate into consideration. The algorithm will identify the trajectory that can't match the trajectory template based on trajectory analysis.