Scribd
Upload
2 views

Research Proposal - Enhancement of CNN-based Object Detection in Autonomous Vehicles

This proposal focuses on enhancing CNN-based object detection in autonomous vehicles to improve safety and efficiency. It outlines three main approaches: fine-tuning CNN architectures, augmenting training datasets for diversity, and optimizing algorithms for real-time performance. The goal is to enable autonomous vehicles to navigate complex environments with greater accuracy and responsiveness.

Uploaded by

Muhammad Khalid
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
2 views

Research Proposal - Enhancement of CNN-based Object Detection in Autonomous Vehicles

This proposal focuses on enhancing CNN-based object detection in autonomous vehicles to improve safety and efficiency. It outlines three main approaches: fine-tuning CNN architectures, augmenting training datasets for diversity, and optimizing algorithms for real-time performance. The goal is to enable autonomous vehicles to navigate complex environments with greater accuracy and responsiveness.

Uploaded by

Muhammad Khalid
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 2

Talha Abid

2210-6043
Proposed Research Area:
Enhancement of CNN-based Object Detection in Autonomous Vehicles

Introduction
Autonomous vehicles have the potential to revolutionize transportation by providing safer and more
efficient journeys. However, the accurate detection and recognition of objects in real-time are crucial
for ensuring the safety and reliability of these vehicles. Autonomous vehicles rely on accurate and
efficient object detection to ensure safe and reliable navigation. Convolutional Neural Networks
(CNNs) have proven to be effective in object detection tasks, but there is a continuous need to
enhance their performance. This proposal aims to improve the performance of CNN-based object
detection in autonomous vehicles, enabling them to perceive and respond to their surroundings with
increased accuracy and efficiency. Following are the approaches proposed for enhancing CNN-based
object detection performance in autonomous vehicles.

Proposed Approach 1: Fine-Tuning and Hyper parameter Optimization:


It is proposed to fine-tune existing CNN architectures specifically for object detection in autonomous
vehicles. By fine-tuning the network’s parameters and optimizing hyper parameters, model can be
tailored to the unique challenges and characteristics of autonomous driving scenarios. This approach
will enhance the model’s ability to detect and classify objects accurately in real-world environments.

Proposed Approach 2: Dataset Augmentation and Diversity:


To enhance the performance of CNN-based object detection, the focus will be on augmenting and
diversifying the training datasets. By generating synthetic data or incorporating data from various
sources and conditions, model can be exposed to a wider range of scenarios and improve its
robustness. This approach will enable the model to generalize better and perform well in different
weather, lighting, and traffic conditions.

Proposed Approach 3: Real-Time Optimization and Hardware Acceleration:


Object detection in autonomous vehicles requires real-time processing capabilities. Therefore,
optimization of the CNN-based object detection algorithms can be prioritized for efficiency without
compromising accuracy. Techniques such as model compression, quantization, and efficient network
architectures will be explored. Additionally, hardware acceleration using specialized hardware will be
leveraged to further improve the performance and real-time responsiveness of the object detection
system.

Conclusion:
Enhancing the performance of CNN-based object detection is crucial for the advancement of
autonomous vehicles. By fine-tuning and optimizing the CNN architectures, diversifying and
augmenting datasets, and optimizing for real-time performance, object detection capabilities of
autonomous vehicles can be significantly improved. This proposal aims to enhance the perception
abilities of autonomous vehicles, enabling them to navigate complex environments with increased
accuracy and efficiency.
Reference Papers:
1. Object Detection in Autonomous Vehicles: Status and Open Challenges-2022

This article explores the current state-of-the-art in object detectors and the challenges involved in
integrating them into autonomous vehicles. Object detection is a computer vision task used in various
applications like surveillance systems, mobile text recognition, and medical diagnosis. It is especially
important for autonomous driving, where vehicles need to perceive their surroundings for safe
driving. Object detection algorithms help identify and locate objects such as pedestrians, vehicles,
signs, and barriers in real-time. Deep learning-based object detectors are crucial for this task.

2. Enhanced Object Detection With Deep Convolutional Neural Networks for Advanced
Driving Assistance-2019

The paper proposed CNN-based visual object detection in Advanced Driver Assistance Systems
(ADAS). It proposes de-convolution and fusion of CNN feature maps, which provide context and
deeper features to improve object detection at lower feature map scales. Additionally, author
addressed the issue of object occlusion by implementing soft non-maximal suppression (NMS) across
object proposals at different feature scales. Through comprehensive experiments conducted on the
KITTI dataset with varying image input sizes, author evaluated the effectiveness of the proposed
CNN enhancements.

3. RRPN: Radar region proposal network for object detection in autonomous vehicles-2019
This paper proposes a Radar based real-time region proposal algorithm for object detection in
autonomous driving vehicles. RRPN generates object proposals by mapping Radar detections to the
image coordinate system and generating pre-defined anchor boxes for each mapped Radar detection
point. The author evaluates the method on NuScenes dataset using the Fast R-CNN object detection
network .
4. Real-Time, Cloud-Based Object Detection for Unmanned Aerial Vehicles-2017
This paper highlights the importance of real-time object detection for Unmanned Aerial Vehicles
(UAVs) in various applications. While Convolutional Neural Networks (CNNs) have shown great
potential for image recognition, their computational demands pose challenges for light weight drones.
To address this, the authors propose offloading computation to an off-board computing cloud, while
maintaining onboard capabilities for low-level object detection and short-term navigation. They
utilize Faster Regions with CNNs (R-CNNs), an advanced algorithm, enabling the detection of
numerous object types in near real-time. This approach aims to strike a balance between
computational efficiency and accurate object detection for UAVs.

You might also like