Design and implementation of object detection and classification system based on deformable template algorithm

This paper proposes a real time system for deformable template based object detection and classification, to segment an object of interest from the stationary complex background. Polygonal templates were defined to characterize basic geometrical shapes, such as square and hexagon. The templates have capability to deform its shape to fit the real object. The model is being fitted to the object using its edge information, which extracted using Canny edge detection algorithm. The deformation of the model matched the object by maximizing likelihood probability density function. Likelihood energy is computed in each model position, dimension, and orientation. To limit the computation, detection of the object is performed by calculating its moment energy and defining searching areas. This will result a faster processing time. The system's result is a description about the shape, size, and orientation of the objects.