Development of an artificial vision system to track piglets in farrowing crates, and methods for automatic estimation of behaviour parameters.

详细信息   

• 作者：Navarro Jover ; Jose Manuel.
• 学历：Doctor
• 年：2004
• 导师：Alcaniz Raya, Mariano
• 毕业院校：Universidad Politecnica de Valencia
• 专业：Agriculture, Animal Culture and Nutrition.;Engineering, Agricultural.;Artificial Intelligence.;Computer Science.
• ISBN：9780493550381
• CBH：3138676
• Country：Spain
• 语种：English
• FileSize：11498762
• Pages：376

文摘

In this work we propose to develop a new system that intends to serve as a tool for behavioral and comfort studies of piglets in farrowing pens. A total of 22.608 zenithal images of the farrowing pen have been captured, with an interval between images of five seconds.;In a first stage, a digital image processing software has been developed that detects the position of each pig. The nine piglets present in the farrowing pen have been coloured with spray paints in their loin and sides with different distant colors. The program processes the captured images, looking for color blobs corresponding to the introduced color standards, returning the positions (column X, row Y) of each of the color blobs in each image. The output is a position data file, that contents, for each analyzed image, the X,Y coordinates of each pig (color). The software first segments the image in the RGB color space, to isolate the color blobs from the rest of the image, and afterwards recognizes the color blobs using other color space more suitable for such purpose. The program has been statistically tested, giving a high percentage (72.5%) of detections, 89.1% of which were correct (the rest correspond with confusion error of the program). When a piglet (or blob) has not been detected in an image (27.5% of total blobs processed), it has been classified as: (1) Piglet not present in the image (hidden under the sow or other piglets): 17.54%; (2) Detection error (piglet present and not detected): 2.36% (3) Color found in many parts of the image, so coordinates are not given: 7.60%.;In a second stage, a set of mathematical analysis have been made from the position data files, obtaining new variables, and establishing methodologies for the study and automatic characterization of piglets behavior and comfort. The purpose is not to extract conclusions about behavior itself, but to establish procedures that could aid as a tool to study piglet's behavior. The behaviours studied have been grouped in four categories: (1) Zoning; (2) Nursing; (3) Huddling; and (4) Activity.