Real Time Speed Estimation

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 7

Real Time speed estimation

and traffic management system using Open


CV
Abstract
For real-time speed tracking of moving vehicles, there have been multiple previous implementations.
Although prediction results achieved are promising, these traditional approaches are still far from
being highly accurate and efficient. This could lead to inaccurate predictions, improper reports, and
higher risks of road accidents, and many people who are at default for violating the traffic rules may
go completely unnoticed. The existing system only detects the number of vehicles and also it is done
with an object detection mechanism. In our proposed system, real-time speed estimation vehicle
counting, and lane management functionalities are added. When an input video is given, the speed of
the vehicles moving will be detected and the lane which the vehicles are travelling is also detected
and tracked. Also, all the details of speed estimation and lane management are updated in a separate
CSV file. The proposed system is trained with real-time datasets and also the proposed system has
high accuracy and efficiency
Introduction
Throughout the last several decades, surveillance cameras have been installed all over the transportation
network to aid in traffic management , ensure the safety of drivers, and spot any anomalies  along with
software for detecting cars is becoming more crucial for smart cities  Many cities have installed
cameras along their streets, but it is impossible for a single person to keep track of all of them at once.
Thus, in order for people to accomplish intelligent traffic management, a friendly intelligent traffic
monitoring system is required.Accurate vehicle detection is the initial step of intelligent traffic
monitoring . Real-time road monitoring, intelligent tracking, and intelligent traffic management are
only some of the many uses for this method . The purpose of our system is, thus, to identify vehicle
categories and detect vehicles in real time.Over the last decade, many vehicle identification standards
with varying degrees of difficulty have been presented. Meanwhile, advances in deep learning-based
techniques have led to remarkable progress in the realm of vehicle detection and object detection areas,
and may be broken down into one-stage and two-stage detection algorithms. The associate editor
responsible for managing the evaluation of this paper and giving final approval was Sudipta Roy.
Module 1
Dataset Collection:

The dataset for the project should be able to identify the speed at which the particular vehicle is travelling from a
video. Using Kaggle, we obtain the dataset and then label them. This labelled dataset is trained using an algorithm for
machine learning to track the vehicle's speed. The dataset consists of different vehicle images with label annotations.

Preprocessing:

Image reshaping, Image resizing , Cropping , Contrasting of Image, Scaling and Augmentation . This all are done
during the Pre processing time.

Splitting the data:


The overall dataset gets split into 2 parts : 
Training 
Testing
Training – Training consists of 70 % of the data set.
Testing – Testing part consists of remaining 30 % of dataset. 
Module 2
• The processed data which gets split into 2 parts , out of it the training data is then fed into that
system’s architecture to predict the output.

• Eventually the model gets trained using the dataset in training period using different Algorithms
and their efficiency gets evaluated . Algorithm’s effieciency and accuracy plays an important role
in determining which algorithm is suitable for the model to predict out put with high efficiency

• RCNN is found to have better accuracy than the other object recognition algorithms
Module 3
• The model is then fine-tuned to recognize lane boundaries and a speed estimate threshold based on
the object recognition training.

• Then the output is displayed and generated in a web application , from the model .

• For users to share their video stream and see their predictions in real time, a web application is
built in which the output data is displayed for users to share.

• A separate.csv file for each group of projections to be generated and for future monitoring
purposes.
System Architecture
Train
Image data
Splitting
Dataset Image preprocessing annotationusin the data
g labeling

Test data
Performance Epochs=100
metrics Checkpoints=true YOLO Algorithm Traning
Evaluation History callbacks

Save Load the Web-App to vechicles


the model with upload the video Input video detection
model keras stream

Vanishing
Speed estimation Background Point and 
Motion ROI
using depth extraction Edge
Tracking  identification
based measures detection
Violation detection Recorded
based on speed Details

You might also like