Our Herbie Family
陳威融
Warner Chen
No. 4
楊沐穎
Remy Yang
No. 7
程俊壹
Joey Cheng
No. 9
張仕嘉
Scott Chang
No. 15
林建帆
Ryan Lin
No. 16
Project Scope
ISSUES
MISSION
- Truck driver shortage
- Truck driver fatigue accidents
To create an autonomous RC truck that can navigate a race course using:
- Image Recognition
- Deep Learning
Donkey Car
Tamiya RC car chassis
PWM
Servo Driver
Raspberry Pi
Camera
Donkey Car
Tamiya RC car chassis
PWM
Servo Driver
Raspberry Pi
Camera
Raspberry Pi
3 board
Softwares
Features
Follow in lane
Automatic follow car
Road sign detection
Traffic light and turn road sign detection
Obstruction detection
Voice Commands
Follow In Lane
Model Training
Data Collection
Test the Model to drive our Car
Step 1
Step 2
Step 3
Crate a model !
林建帆 / No. 16
Track Design
Training Track
Testing Track
Track Design
Noise
Track Lane
Noise
Track Lane
Track Lane
Without Noise
Without Noise
Improve track
Environment
Train a Model
Hi~I am
Model XXX
Data Collection
The Data Sets We Collect
Image file
JSON file
Record 10 data sets per second
The Neural Network
5 Layers CNN
Convolution
Model One Training
10,000 Data Sets
Total :
30,000
Data Sets
10,000 Data Sets
10,000 Data Sets
Model One Demo
Hi~I a.m..
M..od...el.. "P.ro..type.."
Model Two Training
10,000 20,000
Total :
60,000
Data Sets
10,000 20,000
10,000 20,000
Data Sets
Data Sets
Data Sets
Model Two Demo
Hi~I am
Model "Great"
Add OpenCV Function
Model Three Training
20,000 Data Sets
Total :
80,000
Data Sets
Plus 10,000 Data Sets !
Plus 10,000 Data Sets !
20,000 Data Sets
20,000 Data Sets
Model Three Demo
Hi~I am
Model "Genius"
Obstruction Detection
Training Steps
Allocate traffic cones
Drive clock / counter clock wise
Collected 60,000+ photos
Train model
Test model
Demo
Follow
follow the target
with direction
stop when recognize
specific object
{"user/angle": -0.7275541958055998, "user/throttle": 0.5386390125050264}
CNN
picamera
Demo
Traffic Light
Server
Raspberry Pi
photo
Socket
Traffic Light
Server
result
{RED}
Socket
Raspberry Pi
Demo
Left & Right Turn
Server
Raspberry Pi
{Right}
Socket
There are 250~300
labeling pictures
for each traffic sign
Demo
Road Sign Detection
Training step
2000+ pictures & 2000+ XML
Training step
{straight}
(picture)
(XML)
Training step
80%:training
20%:testing
Yolo
Tensorflow
Tensorflow models zoo
F-rcnn
mobilenet
Demo
Transition Period
Voice Commands
Google Web Speech API
Socket
Demo
Traffic Status Chatbot
What's Next?
- Integrate Robot Operating System (ROS)
- Replace Raspberry Pi with a more powerful computer
Thank You
ありがとうございました
謝謝
Final Draft
By warner1
