Semantic segmentation and object detection are critical tasks for enabling self-driving cars to perceive and navigate their surroundings safely.
Semantic segmentation involves labeling each pixel in an image with its corresponding object class. This detailed understanding of the road scene aids in safe navigation and decision-making.
Object detection is a fundamental computer vision task that involves identifying and locating objects of interest within an image. For self-driving cars, object detection helps in identifying various obstacles, pedestrians, other vehicles, and road signs in the environment.
The U-Net architecture is employed for semantic segmentation due to its effectiveness in capturing spatial information. It features a contracting path for context extraction and an expansive path for accurate segmentation.
For object detection, both YOLOv3 and YOLOv8 models are utilized. YOLO (You Only Look Once) models are chosen for their efficiency and accuracy. They divide an image into a grid and predict bounding boxes and class probabilities for objects in each grid cell.
The project utilizes the following datasets from Kaggle:
- For semantic segmentation: Semantic Segmentation for Self Driving Cars.
- For object detection: Lyft 3D Object Detection for Autonomous Vehicles.
- Training accuracy: 98.02%
- Validation accuracy: 97.78%
- Sample Segmentation Results:
- Sample Object Detection Result:
- Sample Object Detection Result:
The code folder contains model codes. Explore more results and the models' performance in the project's Jupyter Notebook.