Object-Detection on Scooters

The objective of the project is to train a pre-trained Yolo model to detect Electric Scooters. For this project, I used YOLOv4 model, which was already trained on the COCO dataset and can detect 80 common objects such as vehicle, bike, and motorcycles; however, YOLOv4 was not trained on electric scooters, so I needed to re-train the model with scooter images so that the model could be used to detect scooters.

Project Highlights

The project consists of the following steps:

• Prepare and pre-process the scooter images
• Download and configure a pre-trained YOLOv4 model
• Configure the Darknet framework to train YOLOv4 model
• Train YOLO with the images of electric scooter
• Perform object detection using the new trained model
• Count the number of scooters in an image

Object-Detection Samples

Example 1

Example 2

Example 3

Image Pre-processing

Sample Distorted Image

Brightening

Denoising and Detection

Technologies

• YOLOv4: the pre-trained weights and configuration file.
• Darknet: an open-source framework for training YOLO models.
• Google Collab: used for running Darknet and training Yolo.
• LabelImg: used for image annotation.
• Python
• OpenCV
• Numpy
• Matplotlib