Calibrated markerless image processing for the detection of object inherent motion

The availability of systems that allow the movements and behavior of different animal species as well as the dynamics of the development of populations as a whole to be recorded and automatically evaluated using objectively collected measured values is of great interest both nationally and internationally. The drivers of this need are the initiatives launched worldwide to preserve biodiversity, such as the European Biodiversity Partnership, as well as the public discourse on the subject of animal welfare, which includes both the area of farm animals as well as the area of luxury animals and animals kept in zoological facilities.

Currently, subjective evaluations and the use of simple technical systems dominate in practice in the context of monitoring tasks in the above-mentioned application areas. Reference should be made to wildlife telemetry, the use of camera traps and the use of different score-based evaluation grids, such as the Numerical Rating Scale of Behavior. In the research area, activities can be demonstrated in various centers at the same time, which address, for example, the latest developments in the area of modeling for predicting the development of biotopes and the use of artificial intelligence methods for the evaluation of sensory-detected data.

The research approach, which is the subject of the validation project and which was originally researched in the computer vision department of the Humboldt University in Berlin, focuses on markerless - and therefore contact-free - visual and very sensitive detection of movement and behavior patterns of animals in different settings. For this purpose, modern methods of object recognition, tracking and analysis of movement patterns are combined with machine learning techniques in order to be able to derive a robust system based on objective measurements for automatic movement/behavior analysis and re-identification of animals.

An advantage of the research approach is the claim to generalization. This means that the models and algorithms researched at HU-Berlin can be applied virtually independently of the specific external appearance of a species. The further investigation of this abstractly defined extraction of movement and the representation or detection of behavioral changes is part of the follow-up activities to be defined, which are to be prepared with the validation project, as part of which, in addition to technical further development in relation to the calibration and recalibration of optical systems Outdoor conditions will focus on the identification and evaluation of different application scenarios that can be implemented in the future.