Comparative Analysis of Number of Person for Dense and Sparse Crowdy Scenario

Abstract

In this paper we presented a comparative analysis of number of person for dense and sparse crowd scenario. Accurate person counting in images is a critical task with numerous applications in various domains such as public safety, urban planning, and retail analytics. However, counting individuals in images poses significant challenges, particularly in scenarios with varying crowd densities. In this dissertation report, we present a comparative analysis of three distinct models for counting the number of persons in both dense and sparse images: regression model, R-CNN (Region-based Convolutional Neural Network), and VGG16 (a deep CNN architecture).

The primary goal is to evaluate the performance of each model in terms of test loss and Mean Absolute Error (MAE) score. We began by providing an overview of the crowd counting task and the challenges associated with it, followed by methodology section, where we described the implementation details of each model and discuss their respective strengths and weaknesses. To conduct our analysis, we collected a diverse dataset of images containing both dense and sparse crowds. We preprocessed the data and annotate it with ground truth labels to facilitate model training and evaluation. We then trained each model using the annotated dataset, employing appropriate optimization techniques and hyper parameter tuning.

Following model training, we calculated the performance of each model using standard metrics like Mean Absolute Error (MAE) and Mean Squared Error (MSE). We also performed qualitative analysis by visually inspecting the model outputs on test images. Through comparative analysis, we identified the model that performs state of art in different crowd density scenarios.