Real-time Accurate Object Counting for Smart Farms

Abstract

Automatic methods that count the number of living bodies is extremely valuable for modern farms. Recently, there are an increasing number of studies conducted on imagery based automatic object counting. However, these studies suffer from extreme overlap of objects, scene perspective, the size of instances and etc. Moreover, real-time performance needs to be met in many application scenarios. In this study, we propose a video imagery based pipeline for real-time accurate object counting on smart farms. Firstly, we introduce the concept - overlap degree of a frame (ODF). Next, an end-to-end frame filtering algorithm based on ODF is established. Then, we propose an object detection and counting algorithm based on Faster region-based convolutional neural networks (Faster RCNN). Finally, the counting results are fused to accurately obtain the number of detected objects. The experimental results demonstrate that our proposed method can automatically and accurately count the living bodies on farms in real-time. In our experiments, the average speed of detection is 40 frames per second, and the total difference is 1.69%. Therefore, our method is considered to be a suitable tool to conduct automatic object counting on smart farms.