Shape-based recognition and classification for common objects - an application in video scene analysis
Zhi Kai Ku, Chee Fei Ng, Siak Wang Khor · 2010
In this paper, a new system that can recognize and classify the common objects found in a video scene into four object classes, which are four legs animal, vehicle, human, and others object classes, is presented. To classify the objects, the shape features of the objects are extracted from the input images of objects in binary silhouette form. Then, the features are applied to the classification algorithm, which consists of two descriptive ratio tests and one shape test to classify the objects into different categories. Firstly, Width to Height Ratio test (WTHR) is used to differentiate between two groups of objects, four legs animal and vehicle in one group, human and others objects in another group. Subsequently, Base to Abdomen Ratio test (BTAR) is used to differentiate between animal and vehicle objects while Shape Boundary Test (SBT) is used to differentiate the human object from others objects. The proposed system is tested with different dataset containing the common objects listed in above with different pose and position, to check for the system performance and system accuracy. 73.33% accuracy is achieved for Animal object class, 86.67% accuracy for Vehicle object class, 93.33% accuracy for Human object class, and 86.67% accuracy for others object class. An overall recognition rate of approximately 86.67% is achieved.