The research group of Computational Intelligence Group (CIG) from the School of Computing at Universidad Politécnica de Madrid (UPM), in collaboration with a veterinary student from Universidad Alfonso X El Sabio and the Department of Ethology from Eötvös Loránd University of Budapest have carried out a research on canine behavior showing that gender, age, context and individual recognition can be identified with a high percentage of success through statistical and computational methods of pattern recognition applied to their barking.
The canine communication has been a research topic in ethology over the last decade. Most of the research has focused on studying how dogs are capable to understand different forms of human communication, for example by displaying gestures and human voice recognition. This joint research between CIG and UPM aimed to understand the acoustic signals obtained from dog barking when they are subjected to certain situations. This research is conducted through the development of a computational system based on statistic modeling that is able to recognize diverse characteristics of the dog (gender, age, individual, situation…).
The experiments were carried out in Budapest with eight dogs -- three males and five females -- Mudi breed, from Hungary usually used as sheep-dogs. Each dog (aged between one and 10) registered 100 barks. A total of 800 barks was obtained by placing the dog in seven different situations:
(a) alone, the owner tied the dog to a tree;
(b) playing with a ball;
(c) fighting, the human pretended to attack the dog's owner;
(d) receiving their food ration;
(f) in the company of a person who was foreign to the dog; and
(g) to get ready to going out with the owner. Each one of the 800 barks was characterized from 29 acoustic measurements.
By using the diverse computational models obtained from the collected data during the experiment, researchers were able to successfully recognize the dog's gender the 85.13% of the time while the age of the dog (recoded as young, adult and old) was classified without mistakes the 80.25% of the time. The task of identifying the situation in which the dog was it was successful the 55.50%, while the recognition (among the eight dogs participating in the study) of the Mudi breed was successful the 67.63% of the time.
This study reveals the biological relevance and richness of the information in dog barking and brings new possibilities in applied research. For example, the assessment of dog behavior is relevant for diverse organizations, therefore to develop a software programme which is able to identify fear, anxiety and levels of aggressiveness in a dog can be a big help.
Cite This Page: