AI-Augmented Monitoring and Management by Image Analysis for Object Detection and Counting

Abstract

Counting the number of objects from images has become an increasingly important topic in different applications, such as crowd counting, cell microscopy image analyses in biomedical imaging, and horticulture monitoring and prediction. Many studies have been working on automatic object counting with Convolutional Neural Networks (CNNs). This research is aimed to shed more light on the applications of deep learning models to count objects in images in different places, such as growing fields, classrooms, streets, etc. We will study how CNN predicts the numbers of objects and measure the accuracy of trained models with different training parameters by using evaluation metrics, mAP, and RMSE. The performance of object detection and counting using a CNN, YOLOv5, will be analyzed. The model will be trained on the Global Wheat Head Detection 2021 dataset for crop counting and COCO dataset for counting of labeled objects. The performance of the optimized model on crowd counting will be tested with pictures taken on the Texas A&M University campus.