Trajectory Agnostic Livestock Counting Through UAV Imaging

Individual object identification is a common problem in visual tracking. This work aims to resolve the detectable instance disappearance problem by estimating probabilities of its placement in the image. This problem can occur both through occlusion and camera movement. Potential solutions can be used in crowd monitoring, visual tracking, traffic monitoring, and agriculture. We present an evaluation of the proposed method based on cattle counting using an Unmanned Aerial Vehicle (UAV) application which is useful for monitoring farms and eliminates one of the many stress factors on farmers. We are able to not only fuse results from multiple frames but also significantly reduce the number of double-counted individuals in the field.