Assignment

The assignment was to develop a real-time object tracking system to detect and monitor the yellow-legged hornet, an invasive species that poses a threat to honeybees. The project aimed to utilize the YOLOv5 model, a deep learning model for object detection, to achieve this goal.

Approach

We, the team of three, approached this project by dividing the tasks into data collection, labeling, modeling, and deployment. We used Selenium to scrape images from Google and Bing, and then manually cleaned the dataset. We labeled the images using Roboflow, ensuring high standards of annotation. We then trained a custom YOLOv5 model in Google Colab and deployed it in a Streamlit application for real-time object detection.

Result

The end result was a successful real-time object tracking system deployed in a Streamlit application. The system allows users to upload images or videos, or use their webcam for real-time detection of yellow-legged hornets. The model demonstrated high accuracy in identifying and tracking the hornets, contributing to the protection of honeybees.

My Input

I was involved in collecting high-quality images of bees and hornets from various sources, including Google Images, Bing, and YouTube videos. I ensured the diversity and relevance of the images to improve the model's performance. I also participated in the labeling process, meticulously annotating the images to create accurate training data. Additionally, I reviewed the labels created by other team members to maintain consistency and quality. I also played a role in training the YOLOv5 model, experimenting with different parameters and configurations to optimize its accuracy and efficiency.

What I learned

Throughout this project, I gained hands-on experience in various aspects of computer vision, including data collection, labeling, model training, and deployment. I learned the importance of a good dataset and quality labeling for the success of an object detection model. I also learned how to deal with challenges such as model bias and deployment issues.