Milestone 3

Alexander Esenwein, Ryan Schwieterman, Joshua Quinto, Elias Anastasopoulos

Alexander Esenwein, Ryan Schwieterman, Joshua Quinto, Elias Anastasopoulos

​

Overview of Milestone 3

Original:

• Begin testing crosswalk detection feature in real situations
• Have hardware components connected
• Fit the vest to house hardware components
• Continue image training/processing with test images

New:

• Have hardware components connected
• Begin the process of fitting the vest with components
• Improve image training process

Have Hardware Components Connected

Jetson successfully posted with custom linux OS

Camera connected to jetson, camera feed accessible from within the OS

​

Hardware components (continued)

Devices connected: Mouse, keyboard, camera, ethernet cable, hdmi output, usb-c power supply.

Fitting the vest with components

Work in progress 3D printed model to house the jetson.

• STL file made in tinkerCAD �(right)
• Ultimaker Cura is slicing software used

3D Models

• Reference .stp 3D model of Jetson is provided by Nvidia (left)
• Our model ready to print In Cura (right)

Improve Image Training Process

Stopped using Google Colab’s free gpus. Now using a new machine with an RTX 3080 to train A.I. models.

​

400 Image model at 55 epochs.

Old training time: 6 hours

New training time: 5 minutes

Improve Image Training Process (continued)

The extra computing power allowed us to significantly increase the amount of epochs per model.

​

Before we were limited to 55 epochs, which took 6 hours to run. On the new machine, we trained a model at 500 epochs in 1 hour and 41 minutes.

55 epochs

500 epochs

Next Milestone

• Finish connecting all pieces to the vest
• Resume crosswalk detection on real crosswalks
• Implementing I/O features to the vest