2012 ©
             Publication
Journal Publication
Title of Article Boundary Detection of Pigs in Pens Based on Adaptive Thresholding Using the Integral Image and Adaptive Partitioning 
Date of Acceptance 12 October 2016 
Journal
     Title of Journal Chiang Mai University Journal of Natural Sciences 
     Standard SCOPUS 
     Institute of Journal Chiang Mai University 
     ISBN/ISSN 1685-1994 
     Volume 16 
     Issue
     Month April-June
     Year of Publication 2017 
     Page
     Abstract Several studies (Guo et al. 2014, 2015; Kashiha et al. 2014; Khoramshahi et al. 2014; Shao and Xin 2008; Tu et al. 2013, 2014; Wang et al. 2008) have investigated about boundary detection, identification and behavior analysis of pigs in the top-view of the pen. The research on the efficiency of boundary detection of pigs is primary for subsequent tasks, such as pig identification. There are several applications that need pig boundary detection, including pig weight estimation (Brandl and Jørgensen 1996; Kashiha et al. 2014; Kongsro 2014; Li et al. 2015; Wang et al. 2008; Wongsriworaphon, Arnonkijpanich, and Pathumnakul 2015), feeding behavior analysis (Bigelow and Houpt 1988; Haer and Merks 1992; Pourmoayed, Nielsen, and Kristensen 2016; Young and Lawrence 1994), and thermal comfort control for group-housed pigs (Shao and Xin 2008). Some current machine vision systems for pig detection use a general method such as Generalized Hough Transform (Kashiha et al. 2014), a threshold intensity value with the exact background color (for example, black color) that differs from pig colors (Wang et al. 2008), or global thresholding (Shao and Xin 2008). However, Generalized Hough Transform method requires the controlled environment such as sufficient light and a clean pen (Kashiha et al. 2014; Shao and Xin 2008; Wang et al. 2008). Unfortunately, such a method cannot be used in real situations because pig farms have a complex scene (e.g., light changes, urine stains, water stains, and manure). One of popular methods for pig boundary detection is background subtraction method based on a Gaussian Mixture Model (GMM) (Guo et al. 2014; Tu et al. 2013), which can operate with the complex scene. However, a disadvantage of this method is that it leads to an intensive computation (Guo et al. 2014; Tu et al. 2013). Guo et al. (2015) showed that their method could detect pigs based on adaptive partitioning and multilevel thresholding segmentation. Their method requires lower computational complexity than the background subtraction method because their method operates only a single frame image. It does not require multiple frames with the background model while the background subtraction method does (Guo et al. 2014). One of challenging open problems is how to detect pigs in the images taken in a complex scene and a high pen density (Kashiha et al. 2014). There are several threshold segmentation methods. Some researchers studied and presented a comparative analysis of entropy and relative entropy thresholding techniques (Chang, Du, and Wang 2006). Kapur et al. (1985) proposed the usage of maximum entropy thresholding. However, the images obtained in our pig pen have a smaller field of view and more complex scene (many illumination changes). Bradley and Roth (2007) studied about adaptive thresholding using the integral image. Their solution is more robust for illumination changes. Additionally, their method is simple and easy to implement. As a preview, we employ this threshold segmentation and use multiple experimental comparisons. Our article proposes a method for pig boundary detection based on Adaptive Thresholding using Integral Image (ATI) and adaptive partitioning (which is the method that separates connected components to each sub-block). Multiple thresholding applies a threshold to each sub-block. To the best of our knowledge, this is the first publication that attempts to detect pig boundaries in a real-world pig pen with complex scenes and high pen density.  
     Keyword Pig boundary detection, Image segmentation, Adaptive partitioning, Adaptive thresholding 
Author
575040021-5 Mr. PRAWIT BUAYAI [Main Author]
Engineering Master's Degree

Reviewing Status มีผู้ประเมินอิสระ 
Status ได้รับการตอบรับให้ตีพิมพ์ 
Level of Publication นานาชาติ 
citation false 
Part of thesis true 
Attach file
Citation 0