Konkuk Univ. Hackaton (Jan 2023) - Park Jonghyuk(Team leader), 윤정훈, 임준하, 이유진
Developing and upgrading ideas that can solve problems that arise in the process of implementing future transportation by establishing a team under the theme of future transportation. (electric vehicles, hydrogen vehicles, autonomous AI, drones, UAM, individual transportation, etc)
Detachable safety module for personal transportation (Electrical scooter)
- Object detection module (Detecting bump, stop sign, child protection zone sign)
- Preventing drunk driving module
- Preventing one-handed driving module
- Used light-weight AI model (ssd-mobilenet-v2)
- Used Colab, Tensorflow to train the AI and transformed into Tensorflow-Lite
- Labeled, 500 photos per class, a total of 2000 photos
- 4 Classes (Bump1, bump2, stop sign, child protection zone sign)
- Augmented image data by arbitrary application of brightness, saturation, contrast, flip, and rotate
- Batch size = 16, Step number = 20,000
- Mounted the trained model on Raspberry Pi
- Total Loss
-
Dataset Zip file : Dataset Zip.
-
Refer to this github : https://github.com/EdjeElectronics/TensorFlow-Lite-Object-Detection-on-Android-and-Raspberry-Pi
- Used Arduino Uno to implement
- Utilized Alcohol detect sensor (MQ-3) and force sensitive sensor (FSR 402)
- Socktet communication between Raspberry Pi and PC using Python
- Server : Raspberry Pi, Client : PC
- Voice was recored by AI using "CLOVA Voice" provided by NAVER
- Operated 3D printer to make case that is detachable to electrical scooter.
- Utilized laser cutter to produce the additional case of Arduino module.
- Made by 3D printer
- Forward direction of scooter is the camera direction
- Alcohol detect sensor(MQ-3), force sensitive sensor (FSR 402)
- LED, buzzer for warning
- bump, stop sign, child protection zone sign
