High throughput livestock monitoring using computer vision

Abstract

Increasing demand for meat products combined with farm labor shortages has resulted in a need to develop new real-time solutions to monitor animals effectively. Extracting valuable information about pigs from video data is a challenging problem. The focus of this thesis is to apply computer vision techniques on animal video feeds to ensure high throughput monitoring of livestock.

We first consider a pig dataset acquired in controlled lab environments and apply state-of-the-art action recognition methods to it. The 'Pig Novelty Preference Behavioral Dataset' is used to train and validate the performance of different models. We obtain an accuracy of more than 93% and a precision of more than 90% for predicting pig behavior. We open-source our code and annotated dataset at https://github.com/AIFARMS/NOR-behavior-recognition

We then consider a pig dataset acquired in farm environments. Significant progress has been made in continuously locating individual pigs using tracking-by-detection methods. However, these methods fail for oblong pens because a single fixed camera does not cover the entire floor at adequate resolution. We address this problem by using multiple cameras, placed such that the visual fields of adjacent cameras overlap, and together they span the entire floor. Avoiding breaks in tracking requires inter-camera handover when a pig crosses from one camera’s view into that of an adjacent camera. We identify the adjacent camera and the shared pig location on the floor at the handover time using inter-view homography. Additionally, we estimate the time spent at the feeder and drinker for each pig based on proximity to the feeder or drinker. For evaluating the detection module, we use the usual average height (precision) of the precision-recall curve; we obtain 99% for Intersection over Union threshold of 50%. We open-source our code and annotated dataset at https://github.com/AIFARMS/multi-camera-pig-tracking