Investigation and application of tracking algorithm on behaviour analysis for weaners housed in two-climate pens

TY - GEN

T1 - Investigation and application of tracking algorithm on behaviour analysis for weaners housed in two-climate pens

AU - Wang, M.

AU - Larsen, Mona Lilian Vestbjerg

AU - Liu, D

AU - Norton, Tomas

N1 - Conference code: 10

PY - 2022/9

Y1 - 2022/9

N2 - Computer vision based methods for tracking animals such as pigs have recently been studied. However, most studies focus on algorithm development and not on its application for behaviour analysis. The current study explores how a developed tracking algorithm could be applied in the monitoring of short-lasting event behaviours in weaners, including locomotor play and aggression, wherein the need to retain identity of the performing pig for the full duration of the behaviour is a key contsraint. The involved videos were collected from weaners housed in two-climate pens, which include a cover to provide a resting place with higher temperature. This type of housing presents a challenge when developing tracking algorithms due to pigs disappearing and reappearing from the cover, which typically happens when they perform locomotor play and aggression. To deal with such a senario it is necessary to engineer features beyond Intersection over Union (IoU) to achieve sufficient re-ID information for tracking the pigs. This study focused on extracting features for short-time animal tracking of pig agression and playing by designing re-ID features and utilizing these in combination with IoU to improve the tracking. Twenty video clips including locomotor play and aggression behaviours were selected to test the tracking algorithm. The tracking percentage per individual was 75.15% and 54.62% and the ID switching frequency per individual was 0.23 and 0.41 for locomotor play and aggression respectively. The results indicate that the algorithm is more applicable to one involving short-lasting behaviour rather than multiple pigs with the case of reappearance.

AB - Computer vision based methods for tracking animals such as pigs have recently been studied. However, most studies focus on algorithm development and not on its application for behaviour analysis. The current study explores how a developed tracking algorithm could be applied in the monitoring of short-lasting event behaviours in weaners, including locomotor play and aggression, wherein the need to retain identity of the performing pig for the full duration of the behaviour is a key contsraint. The involved videos were collected from weaners housed in two-climate pens, which include a cover to provide a resting place with higher temperature. This type of housing presents a challenge when developing tracking algorithms due to pigs disappearing and reappearing from the cover, which typically happens when they perform locomotor play and aggression. To deal with such a senario it is necessary to engineer features beyond Intersection over Union (IoU) to achieve sufficient re-ID information for tracking the pigs. This study focused on extracting features for short-time animal tracking of pig agression and playing by designing re-ID features and utilizing these in combination with IoU to improve the tracking. Twenty video clips including locomotor play and aggression behaviours were selected to test the tracking algorithm. The tracking percentage per individual was 75.15% and 54.62% and the ID switching frequency per individual was 0.23 and 0.41 for locomotor play and aggression respectively. The results indicate that the algorithm is more applicable to one involving short-lasting behaviour rather than multiple pigs with the case of reappearance.

KW - aggression

KW - animal behaviour

KW - animal welfare

KW - locomotor play

KW - tracking

KW - two-climate pens

UR - http://www.scopus.com/inward/record.url?scp=85172309698&partnerID=8YFLogxK

M3 - Article in proceedings

SN - 978-83-965360-0-6

SP - 417

EP - 422

BT - Precision Livestock Farming '22

A2 - Berckmans, M

A2 - Oczak, M

A2 - Iwersen, M

A2 - Wagener, K

PB - Vetmeduni

T2 - 10th European Conference on Precision Livestock Farming

Y2 - 29 August 2022 through 2 September 2022

ER -