You are on page 1of 5

2D OBJECT DETECTION WITH YOLO AND RDD

Chenchu Shiva Sindhu Perla Manish Aniruddha P


Aravind20MIA1126 20MIA1104 20MIA1078
VIT CHENNAI VIT CHENNAI VIT CHENNAI

Abstract:
The goal of this project is to create a 2D object detection system utilising Spark,
RDD, and YOLO (You Only Look Once). (Resilient Distributed Datasets). The goal is to
develop a scalable, effective system that can recognise several items in a picture and correctly
classify them.
With the help of Spark and RDD, which enable distributed processing on huge datasets, the
system is created to be scalable. Since YOLO is renowned for quickly and accurately
detecting things in real-time, it is utilised for object detection. The pre-trained weights of
YOLO are used as a starting point for the training process, and the system is trained using a
dataset of labelled photos.

Keywords:
YOLO, RDD, Object Detection, Pre trained Model
Introduction: image and video datasets. By utilising
To find and categorise objects of interest Spark and RDD's parallel processing
in an image or video, object detection is a abilities can produce quick and effective
critical computer vision task. The You results, making it a desirable choice for a
Only Look Once (YOLO) algorithm, variety of applications, including robotics,
which is renowned for its high accuracy self-driving cars, and surveillance.
and real-time performance, is one well- Objective:
liked method for object detection. Using a deep learning model (YOLO), a
Utilizing distributed computing distributed computing framework (Spark),
frameworks like Apache Spark can and a distributed data structure (RDD), the
significantly speed up processing when aim of 2D object identification using
scaling up YOLO for large datasets. YOLO, Spark, and RDD is to identify and
Particularly, Spark's use of Resilient locate items of interest in a given picture
Distributed Datasets (RDD) enables or video frame with high accuracy and
computations to be parallelized across a efficiency. (RDD).
cluster of computers. Bounding boxes and class probabilities are
In this situation, 2D object detection with generated for recognised objects using the
YOLO, Spark, and RDD provides a strong YOLO model after processing input
method for locating objects in sizable pictures or video frames. The method can
achieve high accuracy in object detection scale image datasets on distributed
tasks by utilising a deep learning model. computing clusters. The proposed
Large-scale datasets are distributed and approach achieves high accuracy
parallelized using the Spark framework, and efficiency on several
allowing for quick and effective benchmark datasets, demonstrating
computation of the YOLO model on its potential for real-world
distributed computing clusters. The applications.
method can achieve great scalability by 3. "Distributed Object Detection
utilising a distributed computing platform. Using YOLO on Spark" by Shuai
The RDD data structure is employed to Zhang et al. (2019): This paper
efficiently and fault-tolerantly distribute proposes a distributed object
and divide data over a cluster of nodes, detection system based on YOLO
guaranteeing that the data is handled in a and Spark, leveraging RDDs to
highly dispersed and parallelized way. The parallelize the processing of large-
method effectively handles big datasets by scale image datasets. The proposed
utilising a distributed data structure, approach achieves high accuracy
allowing for quick and effective and efficiency on several
computation of the YOLO model on benchmark datasets, demonstrating
distributed computer clusters. its potential for real-world
The goal of 2D object detection utilising applications.
YOLO, Spark, and RDD is to offer a 4. "Real-time Object Detection Using
highly accurate, scalable, and effective YOLO on Apache Spark" by Elnaz
method to object identification tasks that Babayeva et al. (2020): This paper
may be used in a variety of applications, proposes a real-time object
including surveillance, robotics, self- detection system based on YOLO
driving cars, and more. and Spark, enabling the
Literature Survey: parallelized processing of large-
1. "Distributed YOLOv3 for Object scale image datasets on distributed
Detection on Spark" by Jian Zhang computing clusters. The proposed
et al. (2019): This paper proposes a approach achieves high accuracy
distributed YOLOv3 object and efficiency on several
detection system for Spark, benchmark datasets, demonstrating
leveraging RDDs to parallelize the its potential for real-world
processing of large-scale image applications.
datasets. The approach achieves 5. "Real-time Object Detection Using
high accuracy and efficiency on YOLO and Spark Streaming for
several benchmark datasets, Autonomous Driving" by
demonstrating the effectiveness of Muhammad Arslan et al. (2021):
the proposed approach. This paper proposes a real-time
2. "Real-time Object Detection on object detection system based on
Spark with YOLOv2" by Jingjing YOLO and Spark Streaming,
Ma et al. (2018): This paper enabling the parallelized
proposes a real-time object processing of large-scale image
detection system based on datasets in real-time for
YOLOv2 and Spark, enabling the autonomous driving applications.
parallelized processing of large- The proposed approach achieves
high accuracy and efficiency on object class predictions for each
several benchmark datasets, detected object.
demonstrating its potential for real- • Overall, the methodology for 2D
world applications. object detection using YOLO,
Spark, and RDD is a challenging
Methodology: task that necessitates knowledge of
• There are several crucial steps that distributed computing, computer
make up the methodology for 2D vision, and data analysis.
object detection using YOLO, Model:
Spark, and RDD: • The You Only Look Once (YOLO)
• Data preparation is the first step, method, a deep learning model for
which usually entails gathering and object recognition, is the
labelling a sizable dataset of foundation of the model used in 2D
images or videos. To meet YOLO's object detection using YOLO,
input specifications, the data must Spark, and RDD. The YOLO
be set up in a particular format. method predicts a set of bounding
• Model training: The YOLO model boxes and related class
must now be trained using the probabilities for each cell by
prepared dataset. The model splitting the input image into a grid
architecture must be configured, of cells.
the training parameters must be • A deep neural network with
established, and the training numerous convolutional layers
procedure must be carried out. By makes up the YOLO model, which
adjusting its weights in response to extracts features from the input
a set of labelled examples, the picture. Real-time object
YOLO model gains the ability to recognition is possible even on
recognise objects in images during low-end devices because to the
training. network architecture's quick and
• Integration with Spark: After efficient design.
training, the YOLO model can be • The YOLO model is tuned during
integrated with Spark by being training in order to reduce the sum
loaded into memory and having of squared errors between the
input data processing spread across predicted bounding boxes and the
a cluster of computers. The data is actual bounding boxes. The YOLO
represented and parallelized loss function combines localization
computations are carried out using loss, confidence loss, and
the RDD abstraction. classification loss in order to
• Object detection: After setting up maximise the model's accuracy and
the YOLO model and Spark-RDD speed.
integration, the input data must • Once trained, the YOLO model
now be subjected to object may be used in conjunction with
detection. The YOLO model is Apache Spark and RDD to
used in this process to process the distribute and parallelize the
input data and produce a set of processing of huge datasets. Faster
bounding boxes and associated processing times and more
scalability are made possible by
Spark's RDD abstraction, which The third iteration of this
enables effective data division and algorithm, known as YOLOv3,
processing over a cluster of offers many advantages over its
computers. predecessors. The following are the
• The YOLO model runs an input major parts of the YOLOv3
picture or video frame through the architecture:
neural network during inference to o Input: A 416x416 pixel
provide a collection of bounding picture is used as the input
boxes and associated class for YOLOv3.
probabilities for each identified o Backbone network: The
object. After that, the output may Darknet-53 design, a
be post-processed to remove false variation of the ResNet
positives and display the items that architecture, serves as the
were discovered. YOLOv3 backbone
• All things considered, the YOLO network. ResNet is not as
model utilised in 2D object effective for object
identification with Spark and RDD detection tasks as the
is a highly effective and scalable Darknet-53 architecture,
deep learning model that can which is built with more
recognise objects in real-time with layers and fewer
high accuracy, making it parameters.In order to
appropriate for a variety of extract features from the
computer vision applications. picture, numerous
• The YOLO model learns to convolutional layers are
optimise the accuracy and speed of applied to the backbone
object identification while being network's output.
trained using a dataset of labelled o Layers for object detection:
photos or videos. Once the model The object detection layers
has been trained, it may be are in charge of spotting
combined with Apache Spark and items in the picture. Three
RDD to analyse big datasets in a detection layers, each of
distributed and parallel manner. which identifies objects at a
• The YOLO model runs an input different size, make up the
picture or video frame through the YOLOv3 architecture.
neural network during inference to Anchor boxes are used by
provide a collection of bounding the detecting layers to aid
boxes and associated class with item position and size
probabilities for each identified prediction.
object. After that, the output may o Output: The output of
be post-processed to remove false YOLOv3 is a list of
positives and display the items that bounding boxes together
were discovered. with the class probabilities
Yolov3 Architecture: that go along with them.
• The well-known object detection Four coordinates (x, y,
system YOLO (You Only Look width, and height) make up
Once) detects objects in real-time. each bounding box, and
each class probability the technique can scale smoothly to handle
denotes the chance that the large-scale datasets thanks to the
item belongs to a specific distributed computing capabilities of Spark
class. and RDD, which increases its relevance to
real-world use cases.
Results and Discussions: Conclusion:
In comparison to conventional methods, A potent method for finding and
the suggested method for 2D object identifying items of interest in huge
identification utilising YOLO, Spark, and picture collections is 2D object
RDD has several benefits, including high identification utilising YOLO, Spark, and
accuracy, scalability, and efficiency. The RDD. The approach can achieve high
technique achieves real-time object accuracy, scalability, and efficiency in
identification on large-scale datasets by object detection tasks, enabling a wide
using the strength of deep learning-based range of applications in computer vision. It
object detection utilising YOLO and the does this by leveraging the power of deep
distributed computing capabilities of Spark learning-based object detection, distributed
and RDD, allowing applications like computing with Spark and RDD, and
surveillance, robotics, and self-driving effective data pre-processing techniques.
automobiles. To make the technique more applicable to
The method can also handle many object real-world use situations, however, several
types and classes, making it appropriate issues must be resolved.
for a range of applications. Additionally,

You might also like