Real-Time Cow Detection and Identification Using Enhanced Particle Filter

In this paper, demonstration of how an individual cow can be tracked and identified in a herd of cows using an enhanced particle filter is presented. In a typical agricultural setup, it is not uncommon to witness cases of cow swapping and rustling. In the past, different methods were recorded in literatures for the monitoring of herds. However, advancement in technology has rendered those methods less effective, resulting to demand for continual improvement on the state-of-the-art methods for tracking individual cows. To achieve this, we employed off-the-shelf networks for end-to-end individual cow identification in up-to-down motionless imagery acquired from stationary camera. The dynamic trajectory of the herd was captured by a camera installed on a pole near a passageway through which the cows passed almost every day and the video processing system was employed for the processing of the herd’s trajectory. Report was made on the model initialisation, mean shift algorithm, particle-Kalman filter, and the target model is updated in order to overcome the change in the target object appearance over time. To our utmost knowledge, this work is the first to apply combination of mean-shift based tracker and particle-Kalman filter as enhanced particle filter to real-time cow detection and identification to curtail occlusion and non-linear tracking problem.