Detection and tracking of pigs in natural environments based on video analysis

Detection and tracking of pigs is important for using computer vision to analyze pig behavior. However, in natural environments, illumination changes, complex scenes, adhesion, occlusion, and individual identification from multiple objects are challenges for detection and tracking. This paper provid...

Full description

Saved in:
Bibliographic Details
Published in:International journal of agricultural and biological engineering 2019-07, Vol.12 (4), p.116-126
Main Authors: Xiao, Deqin, Feng, Aijing, Liu, Jian
Format: Article
Language:English
Subjects:
Adhesion
Agriculture
Algorithms
Automation
Computer vision
Computers
Foraging behavior
Glint
Hogs
Light
Livestock
Methods
Object recognition
Occlusion
Radio frequency identification
Tracking
Video recorders
Citations: Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Detection and tracking of pigs is important for using computer vision to analyze pig behavior. However, in natural environments, illumination changes, complex scenes, adhesion, occlusion, and individual identification from multiple objects are challenges for detection and tracking. This paper provided an anti-interference algorithm for pig detection and tracking based on video analysis. Firstly, pigs were recognized in natural environment based on color information, and noises are removed based on the analysis of connected regions in the binary images. Secondly, multiple pigs are separated by contours and edges. Thirdly, tracking pigs based on a set of association rules with constraint items (DT-ACR). When DT-ACR fails, targets that are not lost were tracked continuously, while lost targets were retrieved in the nearby location, which effectively increase the duration of tracking. Experiments showed that the algorithm was able to track each individual pig in the following conditions: no-light scenes, sun glint scenes, adhesion scenes and occlusion scenes. The overall tracking accuracy reached up to 87.32% {83.85% for serious adhesion, 87.4% for occlusion, 82.4% for strong light, 82.17% for no light and dark, 96.58% for 2 pigs, 88.33% for 3 pigs and 77.63 for 4 pigs). A pig activity analysis study based on the pig detection and tracking algorithm was carried out, and the results showed that our method was able to track pigs for a long period of time and extract the values that reflected pigs' movements.
ISSN:1934-6344
1934-6352
DOI:10.25165/j.ijabe.20191204.4591