International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 05 | May 2024
www.irjet.net
p-ISSN: 2395-0072
Human Activity from Surveillance Camera Using Deep Learning K Sahadevaiah1, Chellaboyina Yaswanth2, 1Professor, Computer Science and Engineering Dept, Jawaharlal Nehru Technological University, Kakinada, AP,
India.
2Post Graduate Student, Master of Technology (IT), Jawaharlal Nehru Technological University, Kakinada, AP,
India. ------------------------------------------------------------------------***-----------------------------------------------------------------------Abstract: In recent years, skeleton-based action recognition has drawn a lot of attention. Because deep learning can extract pertinent information and achieve high recognition accuracy, it has been widely used in picture recognition. Deep learning's ultimate goal is to give machines the same capabilities as the human brain for data analysis and learning by helping them recognise patterns and principles in test data. The goal of this project is to use Media Pipe and deep learning techniques to accomplish robust and accurate human action recognition. The Media Pipe offers pre-trained models that eliminate the need for ongoing training and are useful for accurately identifying important locations on hands, faces, and human bodies. By monitoring their changes over time, these salient points can be utilised as characteristics for action recognition. The convolution and pooling layers of the convolutional neural network get the important information, resulting in an effective action prediction. Based on the input, the model predicts 12 distinct actions. Python is used to implement the deep learning algorithms and media pipe for human action recognition.
not the optimal method for classification; instead, deep learning.
Keywords: Human Activity Recognition, Deep Learning, media pipe, Computer Vision.
These days, deep neural networks are extensively used in both academia and business as the cutting edge of machine learning models in a range of fields, including natural language processing and image analysis.
In the modern world, the most fundamental & effective security measure a building can have is CCTV surveillance. Hospitals, shopping centres, universities, and other establishments use it as the most well-known means of identifying and stopping undesired activity. However, picture an academic campus with over 100 CCTV cameras spread over several structures, such as dorms, classrooms, canteens, sports areas, auditoriums, etc. It is not possible to manually watch every incident captured by the CCTV camera. It takes a lot of time to manually look for the identical incident in the recorded video, even if it has already happened. All things considered, we plan to create a single deep learning model that uses media pipe module data to forecast human behaviour. Lastly, we are contrasting the accuracy using the current ANN technique system. DOMAIN OVERVIEW:
INTRODUCTION
Slowly but surely, these advancements hold great promise for medical imaging technologies, medical data analysis, medical diagnostics, and healthcare overall. We give a brief summary of current developments in machine learning as they relate to medical image processing and analysis, along with some related difficulties. Conventional machine learning techniques were the norm long before deep learning was employed. Like SVM, Logistic Regression, Decision Trees, and Naive Bayes Classifiers.
With the rising of crime rates become an issue if they are not promptly recognised and the appropriate safety measures are not implemented. The majority of cities and metropolitan areas have deployed surveillance systems that continuously gather data. The enormous amount of surveillance data means that there is a greater likelihood of suspicious activity. However, because these jobs are too complex and resource-intensive for artificial intelligence to undertake, human monitoring is necessary to detect such behaviours. One method to simplify an activity for automation is to break it down into smaller components and identify subtasks that could lead to potential crimes. We use our models to try and identify two primary pathways that could lead to crimes.
Another name for these algorithms is flat algorithms. Here, "flat" refers to the fact that these techniques are typically not able to be applied directly to the raw data (text, images,.csv files, etc.). A preprocessing procedure known as feature extraction is required.
OBJECTIVE:
These traditional machine learning algorithms can now employ the representation of the provided raw data as the outcome of feature extraction to complete a task. As an illustration, consider the division of the data into multiple classes or categories.
The primary goal is to use deep learning to construct the model for human action recognition. Probably demonstrating that artificial shallow neural networks are
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 1400