A statistical approach to silhouette tracking

After applying an edge detection step the silhouettes of objects can be detected and tracked through a sequence of images. Traditional methods of object detection ignore information about an object being tracked. This information can be used to increase accuracy and reduce processing time. This paper proposes a method for tracking arbitrarily shaped silhouettes through a sequence of images. It does this by creating a probability distribution function with translation, rotation and scale parameters. It then calculates the expected values of these parameters to detect the object in the image. The processing time and accuracy is compared to a variety of algorithms. It performs for complex object silhouettes.