Animals display a diverse set of behavior, not always well interpreted by humans: some of these behavioral patterns disclose animals’ intensions, needs, and distresses. Thousands of years of animal farming have enabled us to interpret a subset of animal behavior; yet, a large spectrum still remains unknown. Indeed, animal behavioral studies continue to be an active practice, especially due to the role animal plays in working farms and in many industries.
Translating our understanding of animal behavior into an automatic machine algorithm can relieve farmers from time-consuming routine monitoring tasks. For example, tracking animal drinking patterns, sleeping habits, time spent separated from the heard and so forth, can disclose abnormal and/or aggressive behavior as well as enable detection of early symptoms of disease (Animal Health Monitoring). Automatic monitoring of animals is a practical application in the agricultural field that can have great benefits in terms of reducing losses, be it by the way of timely prevention of disease spread or better understanding of any conditions which improve animal productivity and longevity.
Automatic tracking of animal behavior finds its benefits also in research. Deducing symptoms via animal behavior provides an indirect means of assessing the quality of treatment, just like with psychiatric drugs. Deduction of symptoms from behavior analysis can spare invasive dissection and the loss of the animal. In addition, behavioral tests requiring human tracking and measurement allow performing otherwise heavily time-consuming tasks: good examples may be the recording of fish movements and birds in flocks, as well as the extraction of quantitative information out of recording in an efficient and precise manner.
Animal behavior video tracking algorithm
Constructing an algorithm to track animal motion in controlled environment consists of three main steps. In the first step, the animal needs to be detected by algorithmic means. If a study tracks several animals, each of them needs to be found accurately in the frame. In the second stage, a recognition of the detected animal is performed. That is, distinct ID is assigned to each detected animal and tracked over time. The accuracy of the result relies heavily on the ability of the algorithm to correctly track the animal trajectories without confusion. These two steps can be viewed as two classical problems in image processing and computer vision, namely detection and tracking. The last stage consists of high level behavioral inference based on the tracked trajectories. To be able to make usable predictions in regards to the meaning of a trajectory in terms of animal behavior, this last step commonly employs machine learning techniques. The inference is adapted to the characteristic behavior of the tracked animal and inference follows by integrating expert knowledge to automatically distinguish normal from abnormal behavioral patterns and their significance.
To precisely measure how rats learn about new environments, RSIP Vision’s algorithms track the video-recorded movements of the animals’ limbs. Our video tracking animal behavior software does much more than acquire precise information about the rats’ behavior and is now a fundamental tool in the Zoological Department of Tel Aviv University. The software is used to track in real-time how other animals behave. This real time video tracking of behavior is made possible by the algorithm’s ability to predict where the animal will go, allowing for faster processing. Our software produces efficient and accurate research results, freeing researchers to spend their time on analysis rather than on tracking.
RSIP Vision can develop all the solutions you need for tracking and inference by machine learning algorithms. Contact our engineers now to find out how we do that.