Maarten Perneel
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Maarten Perneel has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Jan Verwaeren
Department of Data Analysis and Mathematical Modelling, Ghent University, Coupure Links 653, 9000, Gent, Belgium
3 shared publications Ben Aernouts
Department of Biosystems Division of Animal and Human Health Engineering KU Leuven Campus Geel Kleinhoefstraat 4 2440 Geel Belgium
2 shared publications- Ben Aernouts
Department of Biosystems, Division of Animal and Human Health Engineering, KU Leuven, Campus Geel, Kleinhoefstraat 4, 2440 Geel, Belgium
2 shared publications Ines Adriaens
Department of Data Analysis and Mathematical Modelling Faculty of Bioscience Engineering Ghent University Coupure Links 653 9000 Ghent Belgium
2 shared publications- Ines Adriaens
CRV BV, Global Genetics R&D, 6800 AL Arnhem, the Netherlands; KU Leuven, Animal and Human Health Engineering Division, 2440 Geel, Belgium
2 shared publications Jan Verwaeren
Department of Data Analysis and Mathematical Modelling Faculty of Bioscience Engineering Ghent University Coupure Links 653 9000 Ghent Belgium
2 shared publications
Latest Publications
- Open Access
- 19 May 2025
- Published by Elsevier in Smart Agricultural Technology
- Vol. 11
Consistent multi-animal pose estimation in cattle using dynamic Kalman filter based tracking
- 19 May 2025
- Published by Elsevier in Smart Agricultural Technology
- Vol. 11
Open AccessConsistent multi-animal pose estimation in cattle using dynamic Kalman filter based tracking
- Open Access
Dynamic Multi-Behaviour, Orientation-Invariant Re-Identification of Holstein-Friesian Cattle
To perform reliable animal re-identification, most available algorithms require standardised animal poses. However, this lack of versatility prevents widespread application of these algorithms in behavioural research and commercial environments. To circumvent this, we incorporated information about the orientation and behaviour of the animals in an embedding-based algorithm to re-identify Holstein-Friesian cattle. After all, the orientation and behaviour of an animal determine which body parts of an animal are visible from the camera’s perspective. We evaluated our approach using a dataset with more than 11,000 instance segments of Holstein-Friesian cattle, but our methodology is readily generalisable to different animal species. Our results show that incorporation of informative metadata parameters in the re-identification procedure increases the rank-1 re-identification accuracy from 0.822 to 0.894, corresponding to a 40% reduction in the number of incorrectly identified animals.
Open AccessDynamic Multi-Behaviour, Orientation-Invariant Re-Identification of Holstein-Friesian Cattle
To perform reliable animal re-identification, most available algorithms require standardised animal poses. However, this lack of versatility prevents widespread application of these algorithms in behavioural research and commercial environments. To circumvent this, we incorporated information about the orientation and behaviour of the animals in an embedding-based algorithm to re-identify Holstein-Friesian cattle. After all, the orientation and behaviour of an animal determine which body parts of an animal are visible from the camera’s perspective. We evaluated our approach using a dataset with more than 11,000 instance segments of Holstein-Friesian cattle, but our methodology is readily generalisable to different animal species. Our results show that incorporation of informative metadata parameters in the re-identification procedure increases the rank-1 re-identification accuracy from 0.822 to 0.894, corresponding to a 40% reduction in the number of incorrectly identified animals.
- Open Access
Consistent multi-animal pose estimation in cattle using dynamic Kalman filter based tracking
Over the past decade, studying animal behaviour with the help of computer vision has become more popular. Replacing human observers by computer vision lowers the cost of data collection and therefore allows to collect more extensive datasets. However, the majority of available computer vision algorithms to study animal behaviour is highly tailored towards a single research objective, limiting possibilities for data reuse. In this perspective, pose-estimation in combination with animal tracking offers opportunities to yield a higher level representation capturing both the spatial and temporal component of animal behaviour. Such a higher level representation allows to answer a wide variety of research questions simultaneously, without the need to develop repeatedly tailored computer vision algorithms. In this paper, we therefore first cope with several weaknesses of current pose-estimation algorithms and thereafter introduce KeySORT (Keypoint Simple and Online Realtime Tracking). KeySORT deploys an adaptive Kalman filter to construct tracklets in a bounding-box free manner, significantly improving the temporal consistency of detected keypoints. In this paper, we focus on pose estimation in cattle, but our methodology can easily be generalised to any other animal species. Our test results indicate our algorithm is able to detect up to 80% of the ground truth keypoints with high accuracy, with only a limited drop in performance when daylight recordings are compared to nightvision recordings. Moreover, by using KeySORT to construct skeletons, the temporal consistency of generated keypoint coordinates was largely improved, offering opportunities with regard to automated behaviour monitoring of animals.
Open AccessConsistent multi-animal pose estimation in cattle using dynamic Kalman filter based tracking
Over the past decade, studying animal behaviour with the help of computer vision has become more popular. Replacing human observers by computer vision lowers the cost of data collection and therefore allows to collect more extensive datasets. However, the majority of available computer vision algorithms to study animal behaviour is highly tailored towards a single research objective, limiting possibilities for data reuse. In this perspective, pose-estimation in combination with animal tracking offers opportunities to yield a higher level representation capturing both the spatial and temporal component of animal behaviour. Such a higher level representation allows to answer a wide variety of research questions simultaneously, without the need to develop repeatedly tailored computer vision algorithms. In this paper, we therefore first cope with several weaknesses of current pose-estimation algorithms and thereafter introduce KeySORT (Keypoint Simple and Online Realtime Tracking). KeySORT deploys an adaptive Kalman filter to construct tracklets in a bounding-box free manner, significantly improving the temporal consistency of detected keypoints. In this paper, we focus on pose estimation in cattle, but our methodology can easily be generalised to any other animal species. Our test results indicate our algorithm is able to detect up to 80% of the ground truth keypoints with high accuracy, with only a limited drop in performance when daylight recordings are compared to nightvision recordings. Moreover, by using KeySORT to construct skeletons, the temporal consistency of generated keypoint coordinates was largely improved, offering opportunities with regard to automated behaviour monitoring of animals.
- Open Access
- 1 November 2024
- Published by American Dairy Science Association in Journal of Dairy Science
- Vol. 107 (11) , 9390-9403
Data-driven prediction of dairy cattle lifetime production and its use as a guideline to select surplus youngstock
The lifetime production of dairy cows is a complex trait influenced not only by genetics, but also by the environment in which a cow lives and the management practices of the farmer. Moreover, these influential factors show complex interactions with each other, making it difficult to reliably predict the lifetime production of individual animals at birth. However, since well managed dairy farms often have a surplus of youngstock, reliable lifetime production predictions would offer the opportunity to make more substantiated decisions when selecting calves or heifers to sell. Therefore, using data from Dutch herds, we constructed a data set capturing information on genetics, environment and management practices to develop multiple machine learning models capable of predicting the lifetime production of dairy cattle soon after birth. We found that a coupling of trends observed at the country level with farm-specific models largely outperforms off-the-shelf approaches. At birth, our best model could explain up to 47% of the variance in lifetime production, a considerable improvement in comparison with linear regression on the breeding values supplemented with the average lifetime production at farm level, which could only explain 21.7% of the variance in lifetime production. Moreover, we demonstrated surplus youngstock selection according to our model could more than double the surplus animal selection effect in comparison with the benchmark methodology, offering opportunities to increase the average (future) potential lifetime production of the retained heifers significantly. Assuming a static 20% surplus liveborn heifer scenario and random surplus animal selection as the default, our best model for surplus animal selection resulted in a 9.4% greater lifetime production in the retained animals compared with the current Dutch average lifetime production.
- 1 November 2024
- Published by American Dairy Science Association in Journal of Dairy Science
- Vol. 107 (11) , 9390-9403
Open AccessData-driven prediction of dairy cattle lifetime production and its use as a guideline to select surplus youngstock
The lifetime production of dairy cows is a complex trait influenced not only by genetics, but also by the environment in which a cow lives and the management practices of the farmer. Moreover, these influential factors show complex interactions with each other, making it difficult to reliably predict the lifetime production of individual animals at birth. However, since well managed dairy farms often have a surplus of youngstock, reliable lifetime production predictions would offer the opportunity to make more substantiated decisions when selecting calves or heifers to sell. Therefore, using data from Dutch herds, we constructed a data set capturing information on genetics, environment and management practices to develop multiple machine learning models capable of predicting the lifetime production of dairy cattle soon after birth. We found that a coupling of trends observed at the country level with farm-specific models largely outperforms off-the-shelf approaches. At birth, our best model could explain up to 47% of the variance in lifetime production, a considerable improvement in comparison with linear regression on the breeding values supplemented with the average lifetime production at farm level, which could only explain 21.7% of the variance in lifetime production. Moreover, we demonstrated surplus youngstock selection according to our model could more than double the surplus animal selection effect in comparison with the benchmark methodology, offering opportunities to increase the average (future) potential lifetime production of the retained heifers significantly. Assuming a static 20% surplus liveborn heifer scenario and random surplus animal selection as the default, our best model for surplus animal selection resulted in a 9.4% greater lifetime production in the retained animals compared with the current Dutch average lifetime production.
- Open Access
- 16 January 2024
- Published by Springer Nature in Research Square
Consistent multi-animal pose estimation in cattle using dynamic Kalman filter based tracking
The usage of computer vision to study animal behaviour has gained popularity during the last decade. Replacing human observers by computer vision lowers the cost of data collection and therefore allows to collect more extensive datasets. However, the majority of available computer vision algorithms to study animal behaviour is highly tailored towards a single research objective, limiting possibilities for data reuse. In this perspective, pose-estimation in combination with animal tracking offers opportunities to yield a higher level representation capturing both the spatial and temporal component of animal behaviour. Such a higher level representation allows to answer a wide variety of research questions simultaneously, without the need to develop repeatedly tailored computer vision algorithms. In this paper, we therefore first cope with several weaknesses of current pose-estimation algorithms and thereafter introduce KeySORT (Keypoint Simple and Online Realtime Tracking). KeySORT deploys an adaptive Kalman filter to construct tracklets in a bounding-box free manner and improves the temporal consistency of detected keypoints to a large extent. In this paper, we focus on pose estimation in cattle, but our methodology can easily be generalised to any other animal species. Our test results indicate our algorithm is able to detect up to 80% of the ground truth keypoints with high accuracy, with only a limited drop in performance when daylight recordings are compared to nightvision recordings. Moreover, by using KeySORT to construct skeletons, the temporal consistency of generated keypoint coordinates was largely improved, offering opportunities with regard to automated behaviour monitoring of animals.
- 16 January 2024
- Published by Springer Nature in Research Square
Open AccessConsistent multi-animal pose estimation in cattle using dynamic Kalman filter based tracking
The usage of computer vision to study animal behaviour has gained popularity during the last decade. Replacing human observers by computer vision lowers the cost of data collection and therefore allows to collect more extensive datasets. However, the majority of available computer vision algorithms to study animal behaviour is highly tailored towards a single research objective, limiting possibilities for data reuse. In this perspective, pose-estimation in combination with animal tracking offers opportunities to yield a higher level representation capturing both the spatial and temporal component of animal behaviour. Such a higher level representation allows to answer a wide variety of research questions simultaneously, without the need to develop repeatedly tailored computer vision algorithms. In this paper, we therefore first cope with several weaknesses of current pose-estimation algorithms and thereafter introduce KeySORT (Keypoint Simple and Online Realtime Tracking). KeySORT deploys an adaptive Kalman filter to construct tracklets in a bounding-box free manner and improves the temporal consistency of detected keypoints to a large extent. In this paper, we focus on pose estimation in cattle, but our methodology can easily be generalised to any other animal species. Our test results indicate our algorithm is able to detect up to 80% of the ground truth keypoints with high accuracy, with only a limited drop in performance when daylight recordings are compared to nightvision recordings. Moreover, by using KeySORT to construct skeletons, the temporal consistency of generated keypoint coordinates was largely improved, offering opportunities with regard to automated behaviour monitoring of animals.