Yolov4-deepsort

Yolov4-deepsort is a computer vision system developed to automate the counting of ducks and other livestock animals in a farm environment using live video streams. By combining real-time object detection and tracking, the system eliminates the need for manual counting and provides accurate, scalable monitoring of livestock.

Manually counting a large number of livestock animals in a farm is time-consuming, error-prone, and inefficient, especially in dynamic environments where animals are constantly moving. There was a need for an automated system that could accurately detect, track, and count the livestocks in real-time from live camera feeds without human intervention.

• Real-time Object Detection using YOLOv4
• Multi-object Tracking with DeepSort
• Automated Duck Counting from Live Video Feed
• Custom-Trained Model for Improved Accuracy
• Bounding Box Visualization and Tracking IDs

• Trained a custom YOLOv4 model specifically for duck detection achieving 95% accuracy in farm environments.
• Integrated DeepSort algorithm to maintain consistent tracking IDs across frames for accurate counting.
• Utilized OpenCV for real-time video processing and frame handling.
• Implemented ROI-based object counting, enhancing system automation and data accuracy. .

As the sole developer, I:

• Designed and implemented the complete computer vision pipeline for detection, tracking, and counting.
• Collected and prepared dataset, and trained a custom YOLOv4 model for domain-specific accuracy.
• Integrated YOLOv4 with DeepSort for real-time multi-object tracking and identity preservation.
• Developed logic for automated counting based on tracked object movement across frames.