Study On Eating And Drinking Behaviorsperception Of Caged Chickenbased Onbinocular Vision

China is an egg production and consumption power.And the egg production of China ranks first in the world for 30 years.With the improvement of the nutritional requirements and health awareness of people,the public puts forward higher demand on health status and welfare level of chicken.However,the information acquisition of ill or dead chicken depends on human observation directly,which is tedious,laborious,and inefficient.The behaviors of chicken can be used to indicate their health status and welfare level.But there are few studies of machine vision on the caged chicken to detect their health condition based on eating and drinking behaviors because of similar appearance and large body change between individuals.This research aims to realize the function of eating and drinking behavior perception of caged chicken.The main contents and conclusions of this paper are included below:(1)Building a binocular vision system to realize the function of eating and drinking behaviors perception.The system includes LED,two cameras,a tripod,PC and a software for image capture and image processing.The binocular cameras combined with the software realize the function of image capture,off-line detection,real-time detection,data export and other minor functions.(2)An improved active contour model was proposed to extract the head and body features of caged chicken.Firstly,the rough segmentation result is obtained using the methods of threshold and morphology.Taking the contour of the rough segmentation result as the initial contour of the active contour model,we add open operation of morphology and Gaussian filter into the curve evolution equation to remove the effect of the cage occlusion.The experiment results showed that segmentation accuracy of chicken head was 94.6%using the improved CV model.And segmentation accuracy of chicken body was 91.2%using the improved RSF model.(3)This research realized the function of eating and drinking behaviors perception of caged chicken based on binocular vision system.Multi-threads parallel execution was used to process each cage:Using improved active contour model based on the proposed narrowband building method to acquire the chicken head and body features quickly,combined with the disparity data to construct the small area for 3D reconstruction,then obtaining the 3D contour of the chicken head and body features.Taking the centroid of the 3D contour of the chicken head and body as the reference points,calculating motion parameters and the time length of eating and drinking behaviors.The experimental results showed that the real-time detection of space motion,eating behavior and drinking behavior simultaneously can reach 25 fps.(4)This research studied the relations between the health condition of caged chicken and eating behavior and drinking behaviors.The experimental results showed that ill chicken will decrease eating behavior and drinking behavior.