Scribd Logo
Upload

Welcome to Scribd!

  • Batch 6

    Uploaded by

    Prashanth Kumar
    23 views28 pages
    This document describes a human surveillance system using deep learning that was developed by 5 students from Sir C R Reddy College of Engineering. The system uses YOLO-CNN and LSTM models to detect and track objects in real-time video frames. It was created to address the challenges of continuous surveillance and recognizing complex activities. The document reviews related literature and discusses the proposed system architecture which applies concepts like feature extraction, object detection, and action identification through a CNN, YOLO, and LSTM pipeline.

    Original Description:

    Original Title

    batch6 (1)

    Copyright

    © © All Rights Reserved

    Available Formats

    PPTX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    This document describes a human surveillance system using deep learning that was developed by 5 students from Sir C R Reddy College of Engineering. The system uses YOLO-CNN and LSTM models to detect and track objects in real-time video frames. It was created to address the challenges of continuous surveillance and recognizing complex activities. The document reviews related literature and discusses the proposed system architecture which applies concepts like feature extraction, object detection, and action identification through a CNN, YOLO, and LSTM pipeline.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as pptx, pdf, or txt
    23 views28 pages

    Batch 6

    Uploaded by

    Prashanth Kumar
    This document describes a human surveillance system using deep learning that was developed by 5 students from Sir C R Reddy College of Engineering. The system uses YOLO-CNN and LSTM models to detect and track objects in real-time video frames. It was created to address the challenges of continuous surveillance and recognizing complex activities. The document reviews related literature and discusses the proposed system architecture which applies concepts like feature extraction, object detection, and action identification through a CNN, YOLO, and LSTM pipeline.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as pptx, pdf, or txt
    of 28

    HUMAN SURVEILLANCE

    SYSTEM
    Using DEEP LEARNING
    SIR C R REDDY COLLEGE OF
    ENGINEERING

    DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING


    HUMAN SURVEILLANCE SYSTEM USING DEEP LEARNING

    K.Prashanth kumar 19B81A0593


    K.Ranvitha              19B81A0594
    K.Akash                   19B81A0595
    K.Jayaprakash          19B81A0596
    K.Raviteja                19B81A0597

    Project Guide and Coordinator:


    Dr.K N Madhavi Latha, MTech, Ph.D
    Associate Professors
    Dept. of CSE 
    ABSTRACT

    Human surveillance is the process of recognizing different objects in day to day


    life.

    It is difficult to keep continuous surveillance on any person.

    Complex activities from video are difficult to recognize.

    So, we use deep learning methods YOLO-CNN and LSTM that interrupts with
    activities performed by people and produces relevant label as output.
    PROBLEM STATEMENT
    The system should able to recognize multiple objects

    The recognized data can be used for multiple purposes based on


    user requirements 

    By the end the system would able to recognize single or multiple


    abnormal objects from video.

    Also, it can recognize real-time objects from the videos and from
    you- tube by considering its URL as the input from the user.
    LITERATURE REVIEW 
    Publication details
    Title of article  Description 

    This article is to develop an


    efficient and effective model
    which recognizes the human
    activities from few videos. In this
    paper they mainly used CNN and Ross Girshick, International
    Human Action Recognition using
    CNN and LSTM- RNN with LSTM to identify or recognize the Journal of Innovative
    Attention Model human activities from the Technology and Exploring
    videos.This model will able to Engineering (June 2019).
    identify the simple house activities
    that are been monitored by the
    sensors. For developing this they
    mainly used the LSTM model.
    LITERATURE REVIEW
    Publication details
    Title of article  Description 

    In this article, the model is


    trained using the data that is
    collected using an inertial
    sensor, it is possible to Antonio Bevilacqua, Kyle
    Human Activity Recognition sample some measures such MacDonald, Aamina
    with Convolutional Neural as acceleration and angular Rangerej, Venessa Widjaya,
    Networks velocity of a body, and use Brian Caulfield and Tahar
    them to learn models that are Kechadi(2019).
    capable of correctly
    classifying activities to their
    corresponding classes.
    LITERATURE REVIEW 
    Publication details
    Title of article  Description 

    This article aims to summarize


    various existing abnormal human
    activity recognition (AbHAR)
    A review of state-of-the-art handcrafted and deep approaches Dhiman, Chhavi;      
    techniques for abnormal with the variation of the type of Vishwakarma, Dinesh
    information available such as two-
    human activity recognition dimensional or three-dimensional Kumar (2019)
    data. Features play a vital role in an
    excellent performance of an
    AbHAR system.
    LITERATURE REVIEW 
    Publication details
    Title of article  Description 

    This article provides a method


    that the spatiotemporal features
    of human bones are extracted
    An Intelligent System for through iterative training using
    Detecting Abnormal the Open Pose deep learning Yourong Ding, Ke Bao and
    Behavior in Students Based network and the redundant Jian Zhang, Wuxi Institute of
    information of human bone Technology, Wuxi, Jiangsu,
    on the Human Skeleton and
    facial features is reduced in the China
    Deep Learning feature extraction process,
    effectively reducing the time it
    takes to identify and analyze
    abnormal behavior.
    LITERATURE REVIEW 
    Publication details
    Title of article  Description 

    This article provides a new approach


    to object detection. Prior work on
    object detection repurposes
    classifiers to perform detection.
    Instead, we frame object detection as
    a regression problem to spatially
    You Only Look Once: unified separated bounding boxes and single Joseph Redmon, Santosh
    real time object detection. neural network predicts bounding Divvala, Ali Faradi.
    boxes and class probabilities directly
    from full images in one evaluation.
    Since the whole detection pipeline is
    a single network, it can be optimized
    end-to-end directly on detection
    perfomance 
    EXISTING SYSTEMS
    Faster CNN is an object detection
    algorithm that uses a combination of
    region proposals and a convolutional
    neural network to detect objects. Disadvantages

    RNN is a tracking algorithm that uses a



    Exploit/vanish the temporal correlations of
    series of measurements to estimate the input.
    state of an object over time. It is often
    used in combination with an object

    Frame by frame tracking.
    detection algorithm to track objects as
    they move through a scene.
    PROPOSED SYSTEM
    Advantages
    YOLO-CNN, and LSTM are used to apply
    concepts like feature extraction, object 
    Speed and Accurate
    detection, action detection and identification.

    Detect, deter and report intrusion – when they
    we propose a surveillance system for are about to happening.
    detecting objects and tracking using deep
    learning models .
    .

    Extracting discriminative features
    automatically from the data.


    Keep a backup of important footage securely
    on the cloud.
    System Requirements

    Software:
    Operating System     :        Windows 8 And Above
    Coding Language      :        Python version 3.9
    IDE :        Visual Studio Code and Jupyter

    Hardware:

    Computer System :   Core i5, 4GB RAM And Above


    Other :   Webcam
    SYSTEM ARCHITECTURE 
    METHODS

    CNN

    YOLOV5

    LSTM
    CNN(Convolutional Neural Network)
    CNN is use to extract features form the input video frames.
    YOLO(You Only Look Once)
    YOLO is used to detect objects in each frame.

    YOLO takes the features extracted by the CNN as input.

    It produces the list of bounding boxes.


    LSTM(Long Short Term Memory)

    LSTM is used to analysis the sequence of object detections over time.

    It takes the list of bounding boxes and class probabilities produced by


    YOLO for each frame.
    WORK FLOW 
    DATASET 
    DESIGN

    Use case diagram


    Class diagram
    Sequence diagram
    IMPLEMENTATION
    IMPLEMENTATION
    Input: Output:
    CONCLUSION
     Real-time object detection and object tracking are essential tasks in computer
    vision, with significant practical applications.

     Hybrid methods, which combine deep learning-based and feature-based methods,


    have also been proposed to improve both the detection and tracking accuracy.

     Overall, the choice of approach depends on the specific application requirements,


    such as accuracy, speed, and robustness.
    FUTURE SCOPE

     Retailers can use this information to optimize store layouts, improve customer service,
    and reduce theft.

     This system can help improve patient care and reduce the risk of errors.

     This system can be used to monitor large areas, such as airports or stadiums, and detect
    any suspicious behavior or activity. This can help improve security and prevent crime.
    Thank You

    You might also like