Detecting Poultry Movement for Poultry Behavioral Analysis using The Multi-Object Tracking (MOT) Algorithm

Abstract

Poultry meat is one of the most consumed livestock products in Indonesia. Several studies have concluded that understanding the behavior of poultry will increase production cost efficiency as well as facilitate the fulfillment of animal welfare. Assuring the poultry’s welfare in this increasing business is not an easy task. Using the Multi-Object Tracking algorithm and a pre-trained object detection model trained by the Single Shot Multibox Detector algorithm, we extracted the poultry movement data from a surveillance video with a frame rate of 15 frames per second for behavioral analysis purposes with a precision value of 60.4%. We also managed to gain the object movement plots and periods of the detected objects. This research does not pay attention to the direction of intersecting objects which allows identities to be exchanged between objects. Our near-future research is to add an object identity label in the data preparation step and using a different method of identity assignment which might improve the performance of the algorithm.