EthoVision Multi-Pro is a special version of the EthoVision video tracking system that enables tracking of up to 16 unmarked animals in a single enclosure. During data acquisition the X,Y-coordinates of all objects that are visible in the visual scene are measured and stored on disk. After acquisition, a reconstruction algorithm based on subsequent object positions recreates the movement paths of the individual animals. An initial prototype was presented at Measuring Behavior '96 [1]. Since then, intensive use of this ssystem has resulted in a number of desired extensions to the software, designed to resolve reported problems which are due to the nature of the tracking:

• Animals sitting close together are often detected as a single object.
• Often there are areas in the arena where the animals cannot be detected reliably. Examples of this are hides, insect traps or fodder areas. This results in the disappearance of the animal while it resides in such areas.
• Object size is usually small compared to the arena size: in low contrast areas of the screen this sometimes results in disappearance of the object for small periods of time and in detection of phantom objects.
• Confusion about the identity of objects at the moment that their paths intersect.

Some of these problems can be corrected automatically based on high-level analysis of the animals' paths. Here the concept of path segments is introduced: a path segment is series of object positions that is continuous in time and during which no confusion with other animals occurs. It describes the movement of a single animal over a bout of time. A segment starts when a new object appears and ends when the object disappears. Segments are also separated at positions where confusion between animals can occur.

After the system has automatically detected these segments, several means are available to automatically eliminate the problems mentioned above:

• By discarding segments that are shorter than a preset minimal length, phantom objects are removed.
• The paths of animals that are detected as a single object are restored by duplicating segments.
• By defining hidden zones (areas where the animal can reside without being detected during analysis) it becomes possible to detect (dis)appearance of subjects in animals in hides, on traps, etc.

When the system confuses object identities, correct automatic reconstruction is not always possible. This is the case when multiple animals are located in a hidden zone and one of the animals leaves the zone or when the paths of multiple animals intersect. In these cases, the use of segments allows easy manual editing of the paths. Enhancements to the existing software allow the user to change the identity of an object for all locations within a single segment, linking of separate segments to a continuous path, deleting complete segments and assignment of animals to hidden zones when they are no longer detected.

The presentation will discuss the nature of the problems during tracking and the implemented solutions to these problems. Also, application of these techniques for applications with marked animals will be discussed.

References

1. Buma, M.O.S.; Moskal, J.; Thomas, G.; Jongbloed, S. (1996). Automatic video tracking of multiple animals without the need for marking. Measuring Behavior '96 (Utrecht, 16-18 October 1996), 38-39.