Object detection and video coding using background mosaic plane

Abstract

Object detection in image sequences has a very important role in many applications, such as surveillance systems, tracking and recognition systems, coding systems and so on. We are interested in background subtraction which is very popular algorithm for object detection in image sequences, and also concerned to use the result of detection in selective video coding. Especially when the camera moves and zooms in on something to track the target under drastic illumination change, it``s very hard to detect object properly and the coding efficiency reduces. So, we generate a multiple background mosaic system which is called Background Mosaic Plane (BMP) and use it for object detection and video coding. Some experimental results for both object detection and video coding in various environments show that the averaged performance of the proposed algorithm is good. The research has focused on the following topics. First, we propose BMP structure and generate it to solve the problems for object detection and video coding when there are drastic resolution and illumination change at the same time. BMP is made up of many background mosaics with each different resolution and illumination level which is dynamically calculated and assigned to each mosaic using each descriptor, Resolution Change Descriptor (RCD) and Global Illumination Change Descriptor (GICD). Second, we propose a unified framework for background subtraction, which is made up of three criteria. The two of them are about the used feature and distance metric respectively. The adaptation rule of the background model to illumination changes is considered in the third criterion. We propose an algorithm using spatiotemporal thresholding for object detection with spatio-temporal distance metric. The distance metric is generated by the feature which uses the intensity and gradient at the same time in the feature level instead of in the decision level. In model update process we use Truncated Variable Adaptation Rate ...