Scribd
Upload
64 views

An Application of A Deep Learning Algorithm For Automatic Detection of Unexpected

This document proposes an object detection and tracking system (ODTS) using Faster R-CNN and YOLO models to automatically detect accidents in tunnel CCTV footage. The system is trained on a dataset with cars, people, and fire to achieve average precision scores over 0.7. It then processes video frames over time to detect objects, assign unique IDs for tracking, and identify four types of accidents within 10 seconds of occurring. The system is able to enhance detection capabilities automatically as the training dataset grows without code changes.

Uploaded by

NATIONAL ATTENDENCE
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
64 views

An Application of A Deep Learning Algorithm For Automatic Detection of Unexpected

This document proposes an object detection and tracking system (ODTS) using Faster R-CNN and YOLO models to automatically detect accidents in tunnel CCTV footage. The system is trained on a dataset with cars, people, and fire to achieve average precision scores over 0.7. It then processes video frames over time to detect objects, assign unique IDs for tracking, and identify four types of accidents within 10 seconds of occurring. The system is able to enhance detection capabilities automatically as the training dataset grows without code changes.

Uploaded by

NATIONAL ATTENDENCE
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 7

An application of a deep learning algorithm for automatic detection

of unexpectedaccidents under bad CCTV monitoring conditions in


tunnels

ABSTRACT :
In this project, Object Detection and Tracking System (ODTS) in combination
with a well-known deep learning network, Faster Regional Convolution Neural
Network (Faster R-CNN), for Object Detection and Conventional Object Tracking
algorithm will be introduced and applied for automatic detection and monitoring of
unexpected events on CCTVs in tunnels, which are likely to (1) Wrong-Way
Driving (WWD), (2) Stop, (3) Person out of vehicle in tunnel (4) Fire. ODTS
accepts a video frame in time as an input to obtain Bounding Box (BBox) results
by Object Detection and compares the BBoxs of the current and previous video
frames to assign a unique ID number to each moving and detected object. This
system makes it possible to track a moving object in time, which is not usual to be
achieved in conventional object detection frameworks. A deep learning model in
ODTS was trained with a dataset of event images in tunnels to Average Precision
(AP) values of 0.8479, 0.7161 and 0.9085 for target objects: Car, Person, and Fire,
respectively. Then, based on trained deep learning model, the ODTS based Tunnel
CCTV Accident Detection System was tested using four accident videos which
including each accident. As a result, the system can detect all accidents within 10
seconds. The more important point is that the detection capacity of ODTS could be
enhanced automatically without any changes in the program codes as the training
dataset becomes rich.
EXISTING SYSTEM:
In the existing systemdetects vehicle object and classifies the type of vehicle by
Convolutional Neural Network (CNN). The vehicle object tracking algorithm
tracks the vehicle object by changing the tracking center point according to the
position of the recognized vehicle object on the image. Then, the monitor shows a
localized image like a bird’s viewpoint with the visualized vehicle objects, and the
system calculates the distance between the driving car and the visualized vehicle
objects.

DISADVANTAGES OF EXISTING SYSTEM:


 This system extracts the feature from the satellite image through CNN using
the satellite image as an input value and performs the binary classification
with SVM to detect the vehicle BBox.
 This system utilizes BBox obtained by object detection based on videos or
images. The algorithm applied to the system was compared with the
Gaussian Mixture Model.
 Algorithm: Support Vector Machine (SVM),Convolutional Neural
Network(CNN),Gaussian Mixture Model

PROPOSED SYSTEM:
In the proposed system we attempt is made for generate an object detection &
tracking system (ODTS) with yolo, that can obtain moving information of target
objects with names by combining object tracking algorithm with the deep learning-
based object detection process. It is assumed that ODTS has been trained enough
to perform object detection properly on a given image frame. ODTS receives
selected frames of video at specified time interval c and gains sets of coordinates, n
BBoxs are detected. BBoxT of objects on the given image frame at the time T,
from the trained object detection system. The corresponding type or class ClassT
of each detected object BBoxT is simultaneously classified by the object detection
module.

ADVANTAGES OF PROPOSED SYSTEM:


 A deep learning model of Faster R-CNN was used for training with yolo
object detection model.
 This object tracking module was composed by introducing an object
tracking model called yolo.

Algorithm:R-CNN(Regional Convolution Neural Network),YOLO Model

MODULES:
 User
 Object Detection and Tracking
 RCNN
 Average Precision
MODULES DESCRIPTION:

User:
User can load the cctv videos. To start the project user has to give –input (Video
file path).The open cv class VideoCapture(0) means primary camera of the system,
VideoCapture(1) means secondary camera of the system. VideoCapture(Videfile
path) means with out camera we can load the pre recordedideo file to the system.
After that user has to load the yolo object detection system which is implemented
on RCNN concepts. This yolo module is used for identify the objects from each
frame and name that. It can be idenfiedhumans things fire etc…

Object Detection and Tracking:


Prior detection systems repurpose classifiers or localizers to perform detection.
They apply the model to an image at multiple locations and scales. High scoring
regions of the bounding box of the image are considered detections.We apply a
Regional Convolution neural network to the full image. This network divides the
image into regions and predicts bounding boxes and probabilities for each region.
These bounding boxes are weighted by the predicted probabilities. Our model has
several advantages over classifier-based systems. It looks at the whole image at test
time so its predictions are informed by global context in the image.

RCNN(Regional Convolution Neural Network):


R-CNN models first select several proposed regions from an image (for example,
anchor boxes are one type of selection method) and then label their categories and
bounding boxes (e.g., offsets). Then, they use a CNN to perform forward
computation to extract features from each proposed area. Afterwards, we use the
features of each proposed region to predict their categories and bounding
boxes.Then, based on the detected object information, adependent object tracking
module is initiated to assign theunique ID number to each of the detected objects,
IDt andpredict the next position of each of the objects, BBOX. Thenumber of
tracking BBox u is different from n. But If pasttracked BBox is 0, the number of
tracking BBox equals to thenumber of the detected objects.
Average Precision:
AP values for the target objects to be detected, in the training dataset, the number
of Carobjects is the largest object and very high AP value wasobtained for the Car
object in comparison with other classes.That is, the object detection performance
of deep running ofthe Car in the video was expected to be highly reliable. On the
other hand, AP for Person object results inrelatively low value because Person
object exists long, tinyshape in small size.The AP of Fire object was high, but
falsedetection for the object might be highly possible as the numberof the objects
available for training was very small, Nonetheless, training about deep learning,
including No Fireobjects, could reduce the false detection of Fire object.However,
to detect the Fire in the tunnel control center, it wasnecessary to collect and involve
more images of a Fire eventin training.

SYSTEM REQUIREMENTS:
HARDWARE REQUIREMENTS:

 System : Intel Core i7.


 Hard Disk : 1TB.
 Monitor : 15’’ LED
 Input Devices : Keyboard, Mouse
 Ram : 16GB.

SOFTWARE REQUIREMENTS:

 Operating system : Windows 10.


 Coding Language : Python
 Tool : PyCharm, Visual Studio Code
 Database : SQLite

REFERENCE:

[1] E. S. Lee, W. Choi, D. Kum, “Bird’s eye view localization of surrounding


vehicles :Longitudinal and lateral distance estimation with partial appearance,”
Robotics and Autonomous Systems, 2019, vol. 112, pp. 178-189.
[2] L. Cao, Q. Jiang, M. Cheng, C. Wang, “Robust vehicle detection by combining
deep features with exemplar classification,” Neurocomputing, 2016, vol. 215, pp.
225-231.
[3] A. Arinaldi, J. A. Pradana, A. A. Gurusinga, “Detection and classification of
vehicles for traffic video analytics,” Procedia computer science, 2018, vol. 144, pp.
259-268.
[4] K. B. Lee, H. S. Shin, D. G. Kim, “Development of a deep-learning based
automatic tunnel incident detection system on cctvs,” in Proc. Fourth International
Symposium on Computational Geomechanics, 2018, pp. 140-141.
[5] S. Ren, K. He, R. Girshick, J. Sun, “Faster R-CNN: Towards Real-Time Object
Detection with Region Proposal Networks,” in Proc. Neural Information
Processing Systems, 2015, pp. 91-99.
[6] A. Bewley, Z. Zongyuan, L. Ott, F. Ramos, B. Upcroft, “Simple Online and
Realtime Tracking,” in Proc. IEEE International Conference on Image Processing,
2016, pp. 3464-3468.
[7] K. B. Lee, H. S. Shin, D. G. Kim, “Development of a deep-learning based
automatic tracking of moving vehicles and incident detection processes on
tunnels,” Korean Tunnelling and Underground Space Association, 2018, vol. 20,
no.6, pp. 1161-1175.
[8] C. Dicle, M. Sznaier, and O. Camps, “The way they move: Tracking multiple
targets with similar appearance,” in International Conference on Computer Vision,
2013.

You might also like