International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022
p-ISSN: 2395-0072
www.irjet.net
Locate, Size and Count: Accurately Resolving People in Dense Crowds via Detection Sanyam Swami (Student)1, Prof. Sonal Fatangare (Guide)2, Saisagar Singh(Student)3, Nandakumar Swami(Student)4, Pranay Sankatala(Student)5 1,3,4,5 Student,
Dept of Computer Engineering, RMD Sinhgad School of Engineering, India Dept of Computer Engineering, RMD Sinhgad School of Engineering, India ---------------------------------------------------------------------***--------------------------------------------------------------------2Professor,
Abstract - This way of life makes life easier for people and
Because of the benefits mentioned over, we can train a more robust regression model.
increases the use of public services in metropolises. We present a CNN-MRF- grounded system for counting people in still images from colourful scenes. Crowd viscosity is well represented by the features deduced from the CNN model trained for other computer vision tasks. The neighbouring original counts are explosively identified when using the lapping patches separated strategies. The MRF may use this connection to smooth conterminous original counts for a more accurate overall count. We divide the thick crowd visible image into lapping patches, also prize features from each patch image using a deep convolutional neural network, followed by a fully connected neural network to regress the original patch crowd count. Since the original patches lap, there's a strong connection between the crowd counts of neighbouring patches. We smooth the counting goods of the original patches using this connection and the Markov random field.
2. LITERATURE SURVEY Crowd safety in public places has always been a serious but delicate issue, especially in high-density gathering areas. The higher the crowd level, the easier it is to lose control, which can affect in severe casualties. In order to prop in mitigation and decision-making, it is important to search out an intelligent form of crowd analysis in public areas. Crowd counting and density estimation are precious factors of crowd analysis, since they can help measure the significance of conditioning and give applicable staff with information to prop decision-making. As a result, crowd counting and density estimation have become hot motifs in the security sector, with operations ranging from videotape surveillance to traffic control to public safety and civic planning. A crowd monitoring system is in veritably high demand these days. Still, current crowd monitoring system products have a number of excrescencies, similar as being constrained by operation scenes or having low perfection. In particular, there is a lack of exploration on tracking the number of pedestrians in a large-scale crowded area (see Figure 1). The detectionbased methods and the regression-based methods are the two types of crowd counting styles. Detection-based crowd counting styles generally employ a sliding window to descry each pedestrian in the scene, calculate the pedestrian’s approximate position, and also count the number of pedestrians . For low-density crowd scenes, detection-based methods may produce decent results, but they are oppressively confined for high-density crowd scenes. The early regression based styles attempt to learn a direct mapping between low-level features deduced from original image blocks and head count. Direct regression-based approaches like these only count the number of pedestrians while missing essential spatial information. Learning the linear or non-linear mapping between original block features and their matching target density maps, as indicated by references, may integrate spatial information into the literacy process. Experimenters were inspired by the Convolutional Neural Network’s (CNN) performance in numerous computers vision tasks to use CNN to learn nonlinear functions from crowd images to density maps or counts. In 20205, Wang et al used the Alexnet network structure to
Key Words: Convolutional Neural Network, Sign Language, Machine Learning, Image Processing, Feature Extraction
1. INTRODUCTION There are two major groups of being models for estimating crowd density and counting the crowd direct and circular approaches. The direct approach (also known as object discovery grounded) is grounded on detecting and segmenting each person in a crowd scene to get a total count, while the circular approach (also known as point grounded) takes a picture as a whole and excerpts some features before getting the final count. Due to variations in perspective and scene, the distribution of crowd density in crowded crowd images is infrequently harmonious. As a result, counting the crowd by looking at the entire picture is illogical. As a result, the divide-count-sum approach was acclimated in our system. After dividing the images into patches, a regression model is used to collude the image patch to the original count. Eventually, the accretive number of these patches is used to calculate the global image count. There are two benefits of image segmentation: To begin with, the crowd density in the small picture patches has a fairly invariant distribution. Second, image segmentation improves the quantum of training data available to the regression model.
© 2022, IRJET
|
Impact Factor value: 7.529
|
ISO 9001:2008 Certified Journal
|
Page 3210