Abstract

The field of artificial intelligence (AI) known as computer vision has demonstrated cutting-edge benchmark results in many necessary fields, including object identification, picture classification, image segmentation, etc. Many industrial sectors have benefited as a result of the development of this field of AI. The merger of computer vision with deep learning, also referred to as DL, which is also an area of AI that deals with algorithms inspired by the human brain, has brought machines to a point where machines can easily recognise and detect objects in an image. Convolutional neural networks (CNNs) are the algorithms in deep learning that are responsible for machine vision, which has allowed computers to provide decisions on data based on images.

One of the most valuable aspects of computer vision is object detection, which counts the number of items in an image by creating a bounding box around each one and then counting the bounding boxes. Crowd counting is a crucial application of object detection where the goal is to count the number of objects, especially people, to make it worthwhile for a variety of things like pedestrian counting, depending on how many people entered a market at a specific time, counting the number of people at a specific location at a specific time, and so forth. Applications in this sector are essential regarding security, protecting populated areas from criminal activity, such as walkways and retail malls.

The purpose of this research is to perform analysis and critical evaluation; development of an intelligent system using object detection (OD) techniques to apply the concept of crowd counting, particularly in the context of pedestrians. For applying crowd counting in the context of pedestrians, pre-trained CNN models will be used by applying the concept of transfer learning (TL).

The dataset used in this study will be obtained from Kaggle, an online data science and machine learning platform, and will be used to apply this principle. The model's effectiveness will be assessed by measuring its capacity to draw bounding boxes around images and using loss error. The targets of this research are to critically go through the applications of computer vision, ie, object detection and also to outline new challenges for upcoming researchers aiming to work in this field.

Smart crowd counting using deep learning