◆ Master project using all YOLO versions
Bat Tracking.
Applied vision research with a focus on multi-object tracking in low-light environments.
Computer VisionYOLO BenchmarksResearch
Overview
Applied computer vision project focused on multi-object tracking in low-light environments. The study compares YOLO variants across detection accuracy, tracking stability, and inference efficiency.
- • End-to-end pipeline for dataset preparation and labeling.
- • Benchmarking across YOLO versions and tracking heads.
- • Deployment-ready inference with reproducible metrics.
Project Facts
Focus: Night-vision tracking
Methods: YOLOv5–YOLOv8, DeepSORT
Metrics: mAP, IDF1, FPS
1. Research goals
Establish a reliable baseline for detecting and tracking bats in challenging lighting while preserving high inference speed.
2. Data strategy
Curated a multi-scene dataset with manual labels, augmentation, and temporal consistency checks to reduce false positives.
3. Evaluation
- • mAP@50 and mAP@50-95 for detection quality.
- • IDF1 and HOTA for tracking identity stability.
- • FPS benchmarks on GPU for deployment targets.