Counting and locating high-density objects using convolutional neural network

This paper presents a Convolutional Neural Network (CNN) approach for counting and locating objects in high-density imagery. To the best of our knowledge, this is first object method based on feature map enhancement Multi-Stage Refinement confidence map. The proposed was evaluated two datasets: tree car. For dataset, returned mean absolute error (MAE) 2.05, root-mean-squared (RMSE) 2.87 coefficient determination (R$^2$) 0.986. car dataset (CARPK PUCPR+), superior to state-of-the-art methods. In these datasets, achieved an MAE 4.45 3.16, RMSE 6.18 4.39, R$^2$ 0.975 0.999, respectively. suitable dealing with high object-density, returning performance objects.