DEEP LEARNING BASED CROWD COUNTING AND ANALYSIS

Abstract

Crowd Counting and Analysis is a very important aspect in today’s world, now after the pandemic it has come to light how important crowd management is, as we have to follow social distancing norms that is by reducing the crowd in an area by counting and restricting them. Even to thwart terror threats it has become mandatory for automated lookouts in crowded places for any unscrupulous action by an individual, so that it can be stopped beforehand . The problem because of which crowd scenes is difficult to analyse is due to lot of occlusion, complexity in behavior natures, and posture . Categorized as Crowd Counting and Analysis (CCA) here we aim to achieve the model that can accurately measure the count of persons from an image. I have proposed that a modification in the Multi-Column Convolution Network (MCNN) can further reduce the error in count. Performance analysis on the commonly used datasets are comprehended and cross data validation done. At the end, we deal with problems envisaged and the future aspects of improvement.