• Rishabh Jajoriya
• Pranav Kumar Singh

The three models analyzed in this study - MCNN, CSRNet, and LSC-CNN - are all CNN-based methods that directly estimate the crowd density map from an input image. MCNN uses multiple column networks to capture different scales of crowd information, while CSRNet employs a dilated convolutional neural network to increase the receptive field and better handle scale variations. LSC-CNN, on the other hand, introduces a local self-correction mechanism to reduce the effects of occlusion and background clutter. To evaluate the performance of these models, the study uses three widely used evaluation metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), and the Peak Signal-to-Noise Ratio (PSNR). The outputs tells us that all three models are better than traditional regression-derived methods and best performance is achieved to date on benchmark datasets. However, there are significant differences in their computational efficiency and robustness to variations in crowd density and occlusion. MCNN is the most computationally efficient model but struggles with varying densities and occlusion, while CSRNet is more robust to these factors but has higher computational requirements. LSC-CNN combines the advantages of both models by using a local self-correction mechanism to improve robustness and achieving comparable performance to CSRNet. The study also explores some possible directions for future research, such as combining multiple models or using attention mechanisms to better handle occlusion. Overall, accurate crowd counting is a critical task with numerous practical applications, and CNN-based density map estimation methods have shown significant improvements in recent years. But as we know that there is always possibility for betterment considering efficiency, robustness, and generalization to different scenarios. As such, this study offers valuable insights for researchers and professionals seeking to improve crowd counting techniques and their applications in various domains. By systematically evaluating the strengths and limitations of different models, this study provides a foundation for future developments in this field.

IndexTerms – MCNN, CSRNet, LSC-CNN

"Different cnn models for crowd counting", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.10, Issue 5, page no.a468-a474, May-2023, Available :http://www.jetir.org/papers/JETIR2305064.pdf

"Different cnn models for crowd counting", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.10, Issue 5, page no. ppa468-a474, May-2023, Available at : http://www.jetir.org/papers/JETIR2305064.pdf