Object detection and tracking in video footage play a crucial role in various applications, from surveillance to environmental monitoring. In this project, we developed advanced object tracking and counting algorithms using the DeepSort tracking method, enabling precise enumeration of specific objects within video sequences. Recognizing the importance of accurate data annotation for model training, we collected and meticulously annotated waste data using Roboflow, which was subsequently used to train YOLO models for the detection of different types of waste. Furthermore, to enhance the performance of these YOLO models during both training and inference, we implemented decision-level multi-modality, integrating various data sources to improve accuracy and efficiency. This work contributes to the field of object detection and environmental monitoring by refining algorithmic approaches and optimizing model performance in complex real-world scenarios.

Skills Used

• You Only Look Once (YOLO)
• OpenCV
• RoboFlow
• Algorithm Designs
• Computer Vision

Collaborator

• Ayyan Iqbal

Status

Ongoing