MoodangE_tracking

Install

git clone https://github.com/MoodangE/MoodangE_tracking  # clone
cd MoodangE_tracking
pip install -r requirements.txt  # install

Inference with tracking.py

tracking.py runs inference on a variety of sources, downloading models automatically from the latest YOLOv5 release and saving results to runs/detect.

Command Format

• Save Results as Image Using webcam

  python "file_name".py --source 0 # webcam

• Save Results as Image Using Video file

  python "file_name".py --source "path/vid.mp4" --view-img # video

• Command Format about save results as image using 'custom weights'

  python "file_name".py --weights "path/custom_weight.pt" --data "path/custom_yaml.yaml" --source "path\vide.mp4" --view-img

Test Command (Prediction of Gachon University's Location)

• When collecting information to track your location, it is recommended that you collect it for about 5 seconds.
  Ex) FPS = 30,

1. Predict Location

• Default values

  • Weight : best_1107.pt
  • Data : gachon_road.yaml
  • Start point : AI
  • All files are associated with the realtime database in the firebase to store this information.

• Save Results (Prediction of Gachon University's Location) as Image or Video

  python tracking_location_cam.py --source 0 --view-img

• Save Result (Prediction of Gachon University's Location) as Text

  python tracking_location_text.py  --source 0

• Not Save Result (Prediction of Gachon University's Location)

  python tracking_location_db.py  --source 0

2. Calculate Congestion

• Default values

  • Weight : yolov5s.pt
  • Data : coco128.yaml
  • Filming location : MainGate
  • All files are associated with the realtime database in the firebase to store this information.

• Save Result (Calculate of person confusion) as Image or Video

  python tracking_person_cam.py --source 0

• Not Save Result (Calculate of person confusion)

  python tracking_person_db.py --source 0

Train

1. Change wandb run name in yolov5/train.py

2. Use this command to run training

   cd yolov5
   python train.py --img 1920 --batch 4 --epochs 100 --data customDataset/gachon_road.yaml --cfg models/yolov5s.yaml --weights weights/yolov5s.pt

Implements Reference

• tensorturtle/classy-sort-yolov5 : See How to Configure Two Repositories
• ultralytics/YOLOv5 : With No Modifications
• abewley/SORT : With minor Modifications

Error & Solution

1. PyCharm으로 구동해볼 때 classy-sort-yolov5의 classy_track의 58번째 줄의 **from sort import ***에서 오류가 발생

   해결책 > PyCharm 프로젝트 안에서 두개의 프로젝트를 일반 폴더가 아닌 소스 형태로 바꾸면 정상으로 import 됨

   

   ↓ 정상 import 된 화면

   

2. classy-sort-yolov5/yolov5/weights/download_weights.sh의 파일이 정상 작동하지 않아 yolov5s.pt 파일을 정상적으로 다운로드 하지 못함.

   ** 해결책 >** https://github.com/ultralytics/yolov5/releases/download/v3.1/yolov5s.pt 를 통해 yolov5s.pt 파일을 다운로드 한뒤에 classy-sort-yolov5/yolov5/weights에 붙여넣음

3. (python classy_track.py --source 동영상 경로 --view-img) or (python classy_track.py —source 0)를 실행시 발생하는 오류 ( —view—img: 영상, 0: 웹캠)

   

   AttributeError: 'Upsample' object has no attribute 'recompute_scale_factor’라는 오류가 발생함.

   해결책 > 출처 : ultralytics/yolov5#6948 ~\site-packages\torch\nn\modules\upsampling.py에서 154줄의 recompute_scale_factor=self.recompute_scale_factor 를 주석처리하면 정상 작동

   

   ↓ 정상 작동하며 Multiple Object Tracking (MOT)가 되는것을 볼 수 있음