Automatic monitoring system of sows and piglets during birth and early lactation

Piglet mortality is a problem with complex and multifactorial aetiology, predisposed by the natural biology of the species and exacerbated by economic selection pressures within the pig farming industry. Proper peri- and postpartum management of the sow and the piglets become increasingly important in improving animal welfare and absolutely necessary to reduce the incidence of piglet mortality. There are some obvious areas within the farrowing unit where a stockperson can have positive influence on piglet mortality. On the other hand, interventions performed by farm personnel are often done on a routine basis and not on an evidence-based approach. This means that performed interventions are often redundant and disturb the sows and piglets unnecessarily leading to increased restlessness and nervousness in the animals. Furthermore, peri- and postpartum management is complicated and interventions must be preciously timed and indicated. This is often not possible because stockpersons take care of many animals at the same time. Management of parturient sows and her offspring is very labor-intensive and expensive and there is an increasing societal focus on the ethical and welfare issues in pig reproduction. Further research in reduction in piglet survival must focus on improving management of parturient sows and recognizing factors increasing the risk of piglet mortality in order to meet economic and societal goals.
The major cause of piglet mortality is the starvation-hypothermia-crushing complex. This is a direct result of reduced piglet vitality and reduced maternal behavior. Manual animal behavior observation to assess the health and well-being of animal is nearly impossible particularly in the large scale farms due to the labour and time-intensive nature of the task. Researchers have approached this issue by developing automatic sensor-based systems for monitoring the animal behavior. Computer vision approaches play critical roles in automatic animal monitoring systems and one essential way for measuring animal behavioral effects is pose estimation. The main goal of animal pose estimation is extracting the geometrical configuration of multiple body parts for each individual in order to get information about the behavior of animals. The goal is to provide researchers and farmers with information that can help to understand normal patterns of behavior, e.g., postural changes, maternal behavior, nursing behavior, carefulness behavior, and changes in these patterns, for detecting anomalies, e.g., dangerous behavior of the sow towards piglets.
The main aims of the project were specified and modified by the steering committee as follows:
1) Identification of behavioral risk factors for piglet mortality that can be visualized with a camera system
2) Testing different camera systems and establishing a camera system that works for this purpose
3) Collection of image material and use of open-source data for annotation of key-points of sows and piglets and modelling of multi-pig pose-estimation algorithm for detecting behavior in sows and piglets