Implementing TensorFlow to Assist the Autonomous Agent in Self-Navigating Vehicles
School Name
Governor's School for Science and Mathematics
Grade Level
12th Grade
Presentation Topic
Computer Science
Presentation Type
Mentored
Written Paper Award
2nd Place
Abstract
When it comes to autonomous vehicles, one issue usually brought up is whether a computer can be trusted to navigate a car, something thought only to be operable by humans. With advancements in the field, autonomous vehicles have become the focus of the automotive and computer science fields of research. At the Clemson University International Center for Automotive Research, the project focused on developing the program in an autonomous vehicle in order to recognize traffic signals and signs. This was assisted by the use of a software package named TensorFlow developed by Google. TensorFlow assists the autonomous programming through a process called “Deep Learning”, where the device itself essentially learns what to do in a given situation. Using this application, images were classified into one of six, unique, traffic signals using Inception, a pre-trained model trained on a vast library of pictures. By retraining this model to assist us on our endeavour, a powerful software was developed that could help recognize traffic signs. Developed using the Python language, an algorithm was constructed that classified images according to their function. The images could be recognized at any angle and at a modest distance. We successfully trained the software to work at an accuracy of 60%, after training the code with a relatively low number of images in each category. This accuracy rate suggests that with more training, this program can become more accurate and efficient, thus improving safety in the long run.
Recommended Citation
Rajaraman, Shashaank, "Implementing TensorFlow to Assist the Autonomous Agent in Self-Navigating Vehicles" (2018). South Carolina Junior Academy of Science. 37.
https://scholarexchange.furman.edu/scjas/2018/all/37
Location
Neville 206
Start Date
4-14-2018 9:30 AM
Presentation Format
Oral and Written
Implementing TensorFlow to Assist the Autonomous Agent in Self-Navigating Vehicles
Neville 206
When it comes to autonomous vehicles, one issue usually brought up is whether a computer can be trusted to navigate a car, something thought only to be operable by humans. With advancements in the field, autonomous vehicles have become the focus of the automotive and computer science fields of research. At the Clemson University International Center for Automotive Research, the project focused on developing the program in an autonomous vehicle in order to recognize traffic signals and signs. This was assisted by the use of a software package named TensorFlow developed by Google. TensorFlow assists the autonomous programming through a process called “Deep Learning”, where the device itself essentially learns what to do in a given situation. Using this application, images were classified into one of six, unique, traffic signals using Inception, a pre-trained model trained on a vast library of pictures. By retraining this model to assist us on our endeavour, a powerful software was developed that could help recognize traffic signs. Developed using the Python language, an algorithm was constructed that classified images according to their function. The images could be recognized at any angle and at a modest distance. We successfully trained the software to work at an accuracy of 60%, after training the code with a relatively low number of images in each category. This accuracy rate suggests that with more training, this program can become more accurate and efficient, thus improving safety in the long run.
