| Automatic detection and tracking of moving objects are the fundamental tasks of many video-based surveillance systems. Higher level security assessment and decision making procedures rely upon these essential video analysis tasks. Robust motion detection and object tracking provide the basis for detection of increased activity, entry into a restricted area, detection of objects left behind, tracking of optical flow against established motion patterns, and other similar surveillance requirements. A common method for real-time segmentation of moving regions in image sequences involves modeling each pixel as a mixture of Gaussians, and using K-means approximation to update the model. Each Gaussian distribution represents a background in order to classify a moving object in the adaptive mixture model. This method does not adapt well to graylevel videos given that the color information is limited to a single intensity value. The proposed motion detection and object tracking method is particularly suitable to grayscale videos, such as infrared, thermal, and converted color image sequences. Detection of moving objects in grayscale videos is based on changing texture in parts of the field of view. The proposed method estimates the speed of texture change by measuring the spread of texture vectors in the feature space. It robustly detects very fast and very slow moving objects. The theoretical and experimental results show that measuring spread of texture vectors significantly outperforms Gaussian mixture model methods, such as the Stauffer-Grimson approach. The proposed spatiotemporal motion detection method does not require any post-processing, which is an essential step required by the Gaussian mixture model. Moreover, real-time software designed to detect motion based on the spatiotemporal texture change, is evaluated using color and infrared cameras. The proposed selective hypothesis tracking method is fundamentally based on the location of spatiotemporal texture motion regions, and uses predicted motion vectors, sub-pixel image alignment, and minimum cost estimation of distance, direction, size, and persistence. Motion regions are used to instantaneously update the template of a tracked object. This method is capable of tracking fast and slow moving objects, objects that disappear and later reappear, and divergent and convergent motion regions. |
