Pattern recognition of behavioral events in the nonhuman primate

Abstract

A computerized pattern recognition system has been developed that is capable of identifying 40 separate spontaneously occurring behavioral acts of the primate Macaca fascicularis. The system, called PROBE (pattern recognition of behavioral events), is described in detail. In its present stage of development, PROBE classifies behavioral activity with a reliability comparable to trained human observers. The potential applications for and improvements to the PROBE system are discussed.Techniques used in computer graphics and pattern analysis have been applied to the tasks of observing, classifying, and recording spontaneous behavioral activities in the captive primate. The goal in designing this system was to provide a computer-based pattern recognition system capable of identifying normal activity patterns of the primate Macaca [ascicularis. If successful, this system can be used to detect alterations in these patterns under conditions of toxicologic and pharmacologic insult. In this article, the development of our system, which is called PROBE (pattern recognition of behavioral events), is discussed.Throughout this project, the purpose was to solve a unique application problem. Existing pattern recognition methods (Fu, 1968;Fukunaga, 1972;Pavlidis, 1977), consistent with the adequate solution of our problem, were applied. HARDWAREThe computer system consists of a Digital Equipment Corporation PDP-11/34 computer with a floatingpoint processor and 32K of core memory, a nine-track magnetic tape drive, card reader, line printer, console typewriter, Tektronics Model 4012 display scope, and a 44·million-word disk drive. A schematic view of the

EthoVision: A versatile video tracking system for automation of behavioral experiments

“…In its current implementation, EthoVision is able to detect whether an animal is moving or whether a rodent is rearing, but (in common with other commercial systems) it cannot automatically detect other body postures or behavioral patterns (unless, of course, it is possible to define them in terms of the existing parameters). Research on automatic classification of body postures and behavioral acts in digitized video sequences was already in progress more than two decades ago (Kernan et al, 1980) but has not, to our knowledge, found its way into a commercial product so far. Recently, however, considerable progress has been made with the use of model-based pattern recognition, statistical classification, and neural networks to automatically detect rodent behaviors, such as sitting, grooming, and stretched attend (Rousseau, van Lochem, Gispen, & Spruijt, 2000;van Lochem, Buma, Rousseau, & Noldus, 1998).…”

Monitoring animals’ movements using digitized video images

“…When extended to frame-by-frame playback, where software speed is not as critical, digitized video can also be used for more complex analyses of behavior. For example, items from a repertoire of behavioral fragments can be identified by matching an animal's posture to a series of stereotyped templates (Kernan et al, 1980). Eventually, image-recognition techniques, now much too slow for practical use, may bring the benefits of artificial intelligence to the automated analysis of behavior.…”

An inexpensive microcomputer-based image digitizer

“…Although the system does not have the patternrecognition capabilities of the apparatus described by Kernan, Higby, Hopper, Cunningham, Lloyd, and Reiter (1980), it is effective for collecting data related to animal position and movement at a fraction of the cost. It is unlikely that this system could be programmed to recognize more than the simplest patterns because of the relatively low resolution of the camera and the slow processor speed, which hampers real-time analysis of large data sets.…”