Output.mp4
- Dataset
- Domain: person
- Source: three video yotube with person walking on street (30 fps, total ~85 minute)
- Pre-processing data:
- Extract image in three video (Skip 30 frame, final: ~5000 image)
- Labeling per image using lib LabelImg
- Split data to train model (train & val)
- Test data: ~500 image from Kaggle (follow: https://www.kaggle.com/datasets/constantinwerner/human-detection-dataset)
- Models:
- Using yolov5 custom
- Train model, transfer learning on yolov5s
- Export model to ONNX format & inference by opencv (function ReadNetFromONNX)
- Pipeline IDS:
- Create coords of polygon (Area need to detection)
- In detect, find centroid per object
- Check the centroid with area
- While true: Send massage to telegram user
- Interface:
THANKS FOR WATCHING!