Vision based Obstacle Avoidance for UAVs

Project Title

Vision based Obstacle Avoidance for UAVs

Project Area of Specialization Artificial IntelligenceProject Summary

Object Detection in terms of visual base remains one of the challenging aspects which is still to be sorted out in modern UAVs. There is a need of robust yet accurate trained model for detection of aerial obstacles. We will use SSD-Mobile V2 FPN Lite ML algorithm for the training for Obstacle detection for UAVs. Also, we will use TensorFlow API framework on local machine as well as Google COLAB pro. We will make Six classes for six aerial obstacles. We will train our model around three thousand images. We will also perform performance evaluation of our model using TensorBoard. The data set of around three thousand images in jpg format (around five hundred per class) will be gathered, annotated and then the model will be train successfuly.

Project Objectives

The objectives of this project include:

1. Implementation of Ai technique to build a model which successfuly identifies aerial obstacles for uavs.
2. Reduce human effort and enhance technology use.

Project Implementation Method

We will implement Tensor Flow framework and SSD-Mobile V2 FPN Lite 320x320 to train our model on three thousand images. Firstly Images will be converted to jpg format and then will be annotated using labelImg library, the data then will be split into twenty and eighty percent w.r.t testing and training rewspectively. The dataset will be converted into records for training. Python language will used to build this model. We will perform training on local machine as well as Google Colab Pro. We will then write a detailed thesis report for our work.

Benefits of the Project

Following are the benifits of this project:

1. Successful Identification of aerial obstacles.
2. Reduces human effort
3. Enhances the use of technology,
4. Brings innovation in current Uavs technologies.
5. Lays down paths for further modren research and development.

Technical Details of Final Deliverable

Following are the technical deliverables of this project:

1. Trained Ai Model on three thousand images and six aerial obstacles.
2. Google-Colab Custom object detection training script using Tensor flow.
3. Local machine object detection training script using tensorflow.
4. Custom made dataset of af around seven thousand annotated images for six aerial obstacles.
5. Detailed thesis report of Project.

Final Deliverable of the Project Software SystemCore Industry ITOther Industries Transportation , Media , Security Core Technology Artificial Intelligence(AI)Other Technologies RoboticsSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	37250
Google-COLAB-Pro-Account	Equipment	2	3625	7250
Thesis binding	Miscellaneous	3	2000	6000
Traveling and visits for information gathering	Miscellaneous	2	2000	4000
Graphic Card	Equipment	1	20000	20000