Feature-based orientation estimation for wandering spiders from top-view image sequences

The wolf spiders are important predators of some agricultural pests. They are a natural candidate for a model organism for studying behavioral dynamics, since the population of the wolf spiders is large enough to be easily captured in the field compared to other predatory animals, and since they have a body size suitable for observation of hunting locomotion in the laboratory. This paper presents an image processing algorithm that estimates the orientation of a wolf spider in an observation box from top-view image sequences, where the orientation implies the direction the spider faces. The proposed estimation algorithm first extracts the abdomen-tip of the spider from a captured image, when the spider is motionless. Then, it tracks the abdomen-tip in a given image sequence. The orientation is computed from the abdomen-tip position and the center of gravity of the body in an image coordinate frame, where the center of gravity is used in the sense of image processing, not of physical measurement. An experimental result demonstrates that the orientation estimation algorithm proposed in this paper provides the best performance compared to existing algorithms.