Discover the Cutting-Edge AI Research at NASA

Discover the Cutting-Edge AI Research at NASA

Table of Contents:

1. Introduction
2. Background and Internship Experience
3. The Goal of the Internship Project
4. Step 1: Teaching the Machine to Recognize Airplanes
5. Step 2: Detecting and Localizing Text on Airplanes
6. Step 3: Text Recognition
7. Challenges and Solutions
8. Building a Machine Learning Model
9. Training and Testing the OCR Model
10. Achieving Mission Success
11. Future Steps and Potential Improvements
12. Conclusion

Introduction

In this article, we will Delve into the experience and project of a former NASA intern, Leticia. We will explore her Journey as an intern at NASA and the project she worked on, which involved artificial intelligence and the tracking of aircraft using computer vision and machine learning techniques.

Background and Internship Experience

Leticia begins by introducing herself as a former NASA intern, sharing that she had recently completed a remote internship due to the COVID-19 pandemic. She expresses her openness to questions and proceeds to provide an overview of her internship story.

The Goal of the Internship Project

Leticia explains that the primary goal of her internship was to develop a system capable of recognizing the tail numbers of airplanes at airports. This involved utilizing an Axis camera to track airplanes in the air and implementing computer vision algorithms to detect and recognize the characters on the airplanes' tails. The ultimate aim was to match the recognized tail number with Relevant information about the airplane.

Step 1: Teaching the Machine to Recognize Airplanes

Before the machine could recognize tail numbers, it needed to be trained to identify what an airplane looks like. Leticia used a machine learning platform called TensorFlow to train a model on images of airplanes. She shares examples of the model's performance on different airplane images, highlighting its ability to accurately identify airplanes and the confidence intervals associated with the predictions.

Step 2: Detecting and Localizing Text on Airplanes

Once the machine could recognize airplanes, the next step was to detect and localize the text on the airplanes, specifically the tail numbers. Leticia explains that this task was more complex than object detection and required the use of tools like OpenCV and a deep learning model called EAST. She discusses the challenges faced during this step, such as false positives from detecting text-like shapes in windows, and the techniques used to overcome them.

Step 3: Text Recognition

After successfully detecting and localizing the text on the airplanes, the next challenge was to recognize the actual characters in the tail numbers. Leticia employed a text recognition tool called Tesseract, which utilizes long short-term memory (LSTM) to process the text and determine the individual characters. She showcases the tool's performance on an image of an airplane's tail number, highlighting its ability to accurately recognize the characters, albeit with a few discrepancies.

Challenges and Solutions

Leticia acknowledges that the success of the project was not without its challenges. She discusses the impact of varying orientations and lighting conditions on the machine's ability to recognize tail numbers. To mitigate these issues, she experimented with image pre-processing techniques and explored the possibility of training a custom text recognition model specifically for airplane tail numbers.

Building a Machine Learning Model

Leticia shares her experience of attempting to train her own machine learning model to improve the recognition of tail numbers. She explains the process she followed, including using synthetic and natural images of airplanes, splitting the dataset for training and testing, and the layers of the neural network she built. She acknowledges that the model's accuracy improved quickly initially but did not achieve high accuracy due to limitations in the dataset used for training.

Training and Testing the OCR Model

Despite the limitations of her self-trained model, Leticia emphasizes that the focus of her internship was on learning rather than achieving a perfect final product. She details the different types of images used for training, the challenges faced, and the insights gained from this machine learning endeavor.

Achieving Mission Success

Leticia proudly states that her project achieved mission success by developing a system that could recognize tail numbers off of a video stream of airplanes. She discusses how the project's success was defined and evaluated and underscores the importance of setting measurable goals and prioritizing learning and personal growth.

Future Steps and Potential Improvements

Leticia concludes by outlining potential future steps and improvements for the project. She mentions smart camera tracking, the need for an information hub that connects recognized tail numbers with flight data, and the ultimate implementation of the system at an airport.

Conclusion

In conclusion, Leticia reflects on her NASA internship experience, highlighting the valuable lessons she learned and the joy she found in exploring machine learning and computer vision. She reiterates her willingness to answer questions and invites further inquiries via email.

Highlights

• Leticia shares her experience as a former NASA intern, working on a project involving artificial intelligence and aircraft tracking.
• The goal of the project was to recognize the tail numbers of airplanes using computer vision and machine learning techniques.
• Leticia discusses the challenges faced and the solutions implemented at each step of the project, from teaching the machine to recognize airplanes to detecting and recognizing the tail numbers.
• She explores the process of building a machine learning model and details her attempts to train a model specifically for recognizing tail numbers.
• The project achieved mission success by developing a system capable of recognizing tail numbers off of a video stream of airplanes.
• Leticia outlines potential future steps and improvements for the project, including smart camera tracking and integrating the system into an airport environment.
• The article highlights the importance of continuous learning and personal growth throughout the internship experience.

FAQ

Q: How did Leticia land her NASA internship? A: Leticia initially established connections through her involvement in robotics competitions and a research project with a local professor. This connection led to a recommendation to apply for the NASA internship, which she did and eventually got accepted in her senior year.

Q: Is work still being done on the project Leticia worked on during her internship? A: Leticia mentions that there is ongoing work on the larger project related to aviation. However, with regards to her specific project, she believes it would likely be something she would work on if she were to return.

Q: Did Leticia achieve her goals during the internship? A: Yes, Leticia achieved mission success by developing a system capable of recognizing tail numbers off of a video stream of airplanes. She was able to train and test a text recognition model and obtain accurate results.

Q: What were Leticia's key takeaways from her NASA internship? A: Leticia emphasizes the importance of learning and personal growth during the internship. She highlights the value of defining success in terms of knowledge gained rather than just end results. She also mentions the joy she found in exploring machine learning and computer vision.