[lane detection] traffic sign and traffic light detection utilizing yolov3 and hough transform

term : 2022.05.10 ~ 2022.05.13

flow

1. 객체 인식
   1. data labeling
      • labelImg
   2. yolov3 training
      1. augmentation
      2. use pretrained darknet weights
      3. choose optimizer and loss method (SGD/Adam , bcelogloss/mseloss/cross entropy loss)
      4. optimize hyperparamter (lr, loss weight)
   3. convert yolov3 → darknet’s weights file
   4. convert darknet’s weights → onnx file
   5. onnx file → tensorRT file
2. 차선 인식
   1. utilize detected object
   2. houghline transform
3. path planning
   1. cte ( between detected lane mid position and mid position of image (320 pixel) )

아쉬웠던 점들

path planning

• Pure Pursuit + PID
• Rear-Wheel Feedback + PID
• Front-Wheel Feedback / Stanley + PID
• LQR + PID
• Linear MPC
• https://github.com/zhm-real/MotionPlanning