Lane detection is an important component of autonomous vehicle systems. Highway lane departure warning solutions have been on the market since the mid-1990s. However, improving and generalizing computer vision-based lane detection capabilities remains a challenge. A challenging and difficult task until recently.

CNN - Convolutional Neural Network, also known as the convolutional neural network, is one of the advanced and widely used Deep Learning models in image and video processing problems.

• Input: An image (80, 160, 3).
• Output: A binary mask.

• Convolution
  • Input: A tensor
  • Purpose: Extract feature
  • How it works:
    • An image matrix (volume) of dimension (h x w x d)
    • A filter (f_h x f_w x d)
    • Outputs a volume dimension (h - f_h + 1) x (w - f_w + 1) x 1)
  • Output: A feature map
• MaxPooling2D
  • Input: A tensor
  • Purpose: Reduce input size
  • Output: A tensor
• Dropout
  • Input: A tensor
  • Purpose: Reduce computing
  • Output: A tensor
• Upsample
  • Input: A tensor
  • Purpose: Increase output size
  • Output: A tensor
• Conv2Dtranspose
  • Input: A tensor
  • Purpose: Increase output size
  • Output: A tensor

After 10 times of training with batch size 128, the model ends with MSE for training is 0.0077 and validation is 0.0072.

• Accuracy: 97.79%
• Precision: 98%
• Recall: 89%
• F1-score: 93%
• IoU: 87%

the MSE for training and validation: 0.0059 and 0.0045.

the MSE for training and validation: 0.0052 and 0.0043.

One of the most beautiful roads in Vietnam, it is located next to Phan Thiet beach, Binh Thuan province.

Link input video: from 31:37 to 31:47