In this project, we aim to enhance line thresholding using Convolutional Neural Networks (CNN) and a Multi-Layer Perceptron (MLP) for optimizing Hue, Saturation, and Lightness (HSL) values. The title of this project is "Enhancing Line Thresholding with CNN and MLP Optimization of HSL Values". this project required to the end of the semester about machine learning class with implemented on the car that will help to define which is lane with automaticly or dymic thresholding in order to robust information for the Autonomous car.
For data collection, I designed a custom toolkit and used a robot to capture parameters. The robot was built to assist in data collection, while the toolkit was crafted to help fine-tune and adjust the parameters for optimal performance. Below is the visual representation of both:
| Robot Used for Data Collection | Data Collection Toolkit |
|---|---|
 |
 |
Line thresholding is a critical task in computer vision applications. Our novel approach combines CNNs for feature extraction and MLP for optimizing HSL values. This combination enhances the model's ability to handle variations in lighting and background conditions, leading to more accurate line detection.
The structure of the Deep Neural Network is as follows:
Our DNN architecture consists of multiple convolutional layers that learn hierarchical features from input images. The MLP fine-tunes the line thresholding process by optimizing HSL values. This hybrid approach achieves superior results compared to traditional thresholding methods.
Below are the results from the evaluation of our model and the prediction output from the CNN model:
| MSE Evaluation Result | Prediction Output |
|---|---|
 |
 |
Our CNN model accurately detects lines and edges in real-world images, making it suitable for various computer vision applications, such as image segmentation and object recognition.
Below are the video results of the line thresholding process:
| Original Video | Result Video |
|---|---|
 |
 |