🔬 Research Contribution: Multi-Object Tracking for Precision Poultry Farming

Role: Research Assistant
Institution: University of Georgia
Project Title: Enhancing Multi-Object Tracking of Broiler Chickens using Deep Learning, Machine Learning, and Computer Vision

---

🧠 Overview

Contributed to the development of a robust, real-time, identity-preserving AI tracking system for broiler chickens in commercial poultry farms. The goal was to improve behavior analysis, tracking reliability, and animal welfare using modern deep learning and ML pipelines.

---

🚀 Technical Highlights

1. Object Detection & Optimization

• Trained and benchmarked 10 YOLO variants
• Best model: YOLOv11x
  • Precision: 0.968
  • Recall: 0.960
  • mAP@50: 0.986
  • mAP@50–95: 0.805
• Applied L1 unstructured pruning for latency reduction
  • Inference Speed: Improved from 46.5 FPS → 60 FPS
  • Pruning Ratio: 0.09

---

2. Deep Feature Extraction & Re-Identification

Designed a hybrid deep feature extractor using:

• Vision Transformer (ViT)
• ResNet152
• DenseNet201

Embedding Evaluation Metrics:

• Cosine Similarity: 0.956 ± 0.032
• Euclidean Distance: 0.020 ± 0.007

---

3. Kinematics-Aware Identity Classification

Developed classifiers using features like velocity, acceleration, and displacement. Benchmarked 15 ML models, including:

• Logistic Regression, Random Forest, Extra Trees Classifier (Best)
• Gradient Boosting, XGBoost, LightGBM, CatBoost, AdaBoost
• K-Nearest Neighbors (KNN), Support Vector Machine (SVM)
• Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA)
• Decision Tree, Naive Bayes, Multilayer Perceptron (MLP)

Top Performer: Extra Trees Classifier

• Accuracy: 0.917
• Precision: 0.958
• Recall: 0.920
• F1 Score: 0.939

---

4. Multi-Object Tracking System

Evaluated and optimized 6 tracking algorithms:

• DeepSORT, StrongSORT, SMILEtrack, OC-SORT, ByteTrack, Modified ByteTrack

Final Pipeline Metrics:

• MOTA: 0.904 ± 0.073
• MOTP: 0.953 ± 0.057
• Tracking Speed: 30.1 ± 3.3 FPS
• Continuous Duration: Up to 17.3 minutes

---

📈 Impact & Deployment

Tracked over 5,700 broiler chickens under diverse real-world conditions including:

• Lighting variability
• Occlusions
• Region-specific zones (feeder, drinker, open floor)

Enabled:

• Long-term identity preservation
• Automated behavior monitoring
• Precision livestock farming integrations

This project bridged Computer Vision, ML, and Precision Agriculture, delivering a high-accuracy, scalable pipeline to advance smart farming and animal welfare monitoring systems.