A Study On Pig Object Detection And Tracking In Image Sequences

With the rising demand for pork in recent society, improving the situation ofpig breeding is imperative. In order to judge the growth and health status of pigin sty, behavior of pig is detected. Based on the machine vision and other relevanttechnology, the image acquisition system is established, object detection andtracking ofpig is further studied.All of thecontents in thepaperprovide a methodand foundation to reach the aim of pig behavior detection.The principal contents of the paper are as follows.(1) The background and significance of pig behavior detection based onmachine vision technology is expounded. development situation of machinevision technology applied in animal behavior detection is summarized. The keytechnology of animal behavior detection, such as object detection and objecttracking, is analyzed.(2) The hardware and software of image acquisition is designed. The DH-SV1421GC color industrial camera and the Computar zoom industrial lens arechose as the major equipment of image acquisition. For the automatic acquisition and storage of image data, acquisition software is developed in the environmentofVisualStudio.Inconsiderationofthelargecalculationofsubsequentalgorithm,Gaussian pyramid is applied to compress the scale of image data, which canreduce image distinguishability and maintain image details.(3) In order to detect object in sty, common object detection methods areanalyzed in the aspects of theory, advantage and disadvantage. By studying theapplication of Gaussian mixture model (GMM) and experimenting its flaw indealing with slow-moving or still object, the pig object detection method fusingGMM and image granulating is proposed. Allow for the efficiency and accuracyof image granulating, a series of image processing steps are proposed. SRAD isapplied for image filtering to smooth detail and preserve edge.After the analyzingof common edge detection algorithms, the edge detection algorithm based on fastsegmentation for major edge enhancing is proposed. The marker-controlledwatershed algorithm is applied to finish image granulating. By the analyzing ofGMM result and granulated image, the number of pig can be confirmed, andobject detection is finished with the image fusion of GMM result image andgranulated image. The feasibility and advantage of the proposed method areverified by the experiments on MATLAB.(4)Amethod of pig trajectory tracking based on frame by frame detection ispresented. The method detects object in each frame, and uses the object centroidto replace the object location. Considering the situation of one pig, the methodconnects the object centroid in each adjacent frame by time order, then comes to the trajectory track of object. In the situation of multi pigs, the method connectstheobject centroid in each adjacent frame, combines the nearestneighbor method,then comes to the trajectory track. When there are at least two pigs in pigpen, theobjecttrajectorycanbeobtainedbyusingthenearestneighbormethodandlinkingtarget centroid of adjacent frames. When pigs interfere with each other, the simplenearest neighbor method occurred error. In this paper, a tracing method combinesparticlefilteranddetectionframebyframe hasbeenproposed.Theobjectpositionofpigis estimated byparticle filter in imagesequence.Theassociationofcentroidpoint has been determined by the judgment of nearest neighbor method in nextframe and the basis of estimated position. Finally, the target tracing can beaccomplished by mutual interference of multiple objects.