Research PaperMulti-object extraction from topview group-housed pig images based on adaptive partitioning and multilevel thresholding segmentation
Introduction
The use of monitoring devices on farms has the potential to improve animal welfare (Botreau, Veissier, Butterworth, Bracke, & Keeling, 2007) and raise production levels (Lauber, 2007). With the continuously falling prices and improved performance-price ratio of these monitoring devices, a configuration of monitoring devices in farms becomes possible (Ahrendt et al., 2011, Hu and Xin, 2000). A breeder that uses the current technique of monitoring multiple scenes on a screen for a long period will easily become tired and inefficient. The study of animal behaviour using machine vision technology has become an important and emerging research area.
Research on video monitoring and behaviour analysis of topview group-housed pigs has been given increasing attention (Lind et al., 2005, McFarlane and Schofield, 1995). Monitoring behaviour of pigs in a pen is possible both in groups and at an individual level. Individual level data analysis, however, has more advantages (Kashiha, Bahr, Ott et al., 2013). For individual pig behaviour analysis, the accurate extraction of individual pigs is the first step. The object extraction of a video that contains only a single pig can be obtained effectively using methods such as background subtraction or binarization (Shao and Xin, 2008, Wang et al., 2008). However, for video sequences of topview group-housed pigs with complicated backgrounds (such as those with light changes; urine stains, water stains, manure, and other objects on the ground; slow pig movement patterns; and varying colours of foreground objects), effective object extraction is still challenging and should be further researched. The higher the pen density, the more difficult segmenting the pigs in the image will be (Kashiha, Bahr, Ott, Moons, Niewold, Tuyttens et al., 2014).
For example, in order to detect the position of piglets, Navarro-Jover extracted a piglet from the background using markings on the piglet's back and sides made from different colour spray paints (Navarro-Jover et al., 2009). In order to continuously monitor water use in a group of pigs in a farm pen, Kashiha extracted the pigs using image binarization and segmentation methods (Kashiha, Bahr, Haredasht et al., 2013). Tu, Karstoft, Pedersen, and Jørgensen (2013) proposed a foreground detection algorithm based on loopy belief propagation. This method can overcome the influence of sudden light changes, dynamic backgrounds, and motionless foreground objects. Kashiha, Bahr, Ott, Moons, Niewold, Tuyttens et al. (2014) automatically monitored pig locomotion through image analysis methods such as binarization, morphological processing, and ellipse fitting.
In this paper, a multilevel thresholding segmentation method is proposed to accurately obtain topview multiple pig objects in the drinker and feeder zone, which is a complex scene. The initial segmentation result is obtained through a maximum entropy global threshold segmentation method, and then each object centroid is calculated from the segmentation objects. The radius of each circular sub-block is obtained from the distance curve from the centroid to the edge point. Thus, the original frame is adaptively divided into multiple circular areas. An accurate secondary segmentation result is then obtained through the multilevel thresholding and maximum entropy threshold segmentation within the sub-blocks, which accurately extracts multiple pig objects. The extracted pig individuals can be used to analyse the behaviour of pig drinking and eating as well as to estimate pig weight. We also analysed the feasibility of applying the proposed method to individual pig recognition.
Section snippets
Video acquisition
The experiments for this study were carried out at Danyang Rongxin farm, the experimental base for key disciplines in the Department of Agricultural Electrification and Automation of Jiangsu University. The pigs were monitored in a reconstructed experimental pen. The pen was 1 m high, 3.5 m long, and 3 m wide. A camera was located above the pen at the height of 3 m relative to the ground. The camera used was a Sony FL3-U3-88S2C-C with an image resolution of 1760 × 1840 pixels. Part of the
Experimental results and analysis
We use the original frame shown in Fig. 2(a) as an example. Figure 4(a) is the maximum entropy threshold segmentation result, Fig. 4(b) is the result after extracting the valid area from Fig. 4(a), and Fig. 4(c) shows the result after the morphological processing of Fig. 4(b). The experimental results show that the proposed method can overcome the influence of ground urine stains, water stains, manure, and other background objects using the set of morphological processing steps presented above (
Conclusion
In this paper, a feasible multilevel thresholding segmentation method to accurately obtain pig individuals in the drinker and feeder zone was proposed. The method proposed is suitable for topview group-housed pigs in complex scenes, regardless of pig numbers, colour, slow movements, or behaviour patterns.
Using a coarse segmentation result, each object centroid becomes a bounding sub-block's origin, and the centroid-edge point maximum distance becomes its radius. The original image is then
Acknowledgements
This work was part of a project funded by the “National Natural Science Foundation of China” (grant number: 31172243), the “Doctoral Program of the Ministry of Education of China” (grant number: 2010322711007), the “Priority Academic Program Development of Jiangsu Higher Education Institutions” and the “Graduate Student Scientific Research Innovation Projects of Jiangsu Ordinary University” (grant number: CXLX13_664).
References (28)
- et al.
Development of a real-time computer vision system for tracking loose-housed pigs
Computers and Electronics in Agriculture
(2011) - et al.
Multilevel image segmentation with adaptive image context based thresholding
Applied Soft Computing
(2011) - et al.
Foreground detection of group-housed pigs based on the combination of mixture of Gaussians using prediction mechanism and threshold segmentation
Biosystems Engineering
(2014) - et al.
The automatic monitoring of pigs water use by cameras
Computers and Electronics in Agriculture
(2013) - et al.
Automatic identification of marked pigs in a pen using image pattern recognition
Computers and Electronics in Agriculture
(2013) - et al.
Automatic weight estimation of individual pigs using image analysis
Computers and Electronics in Agriculture
(2014) - et al.
Automatic monitoring of pig locomotion using image analysis
Livestock Science
(2014) - et al.
Validation of a digital video tracking system for recording pig locomotor behavior
Journal of Neuroscience Methods
(2005) - et al.
Review: quantifying animal feeding behaviour with a focus on pigs
Physiology & Behavior
(2015) - et al.
A review on image segmentation techniques
Pattern Recognition
(1993)