FootballTeamDetection_YOLOnOpenCV

Implementation of YOLOv3 with OpenCV for color based classification of objects. (Demonstrated for football team detection)

Requirements

• imutils
• NumPy
• OpenCV

Demo Result

For input video ~1:20-~1:30 snapshots of the results are as shown below:

How to use?

• Clone repository

git clone https://github.com/VenissaCarolQuadros/FootballTeamDetection_YOLOnOpenCV`

• Install requirements if not installed (use a virtual environment if desired)

pip install -r requirements.txt

• Replace yolov3.weights in yolo-coco with the file downloaded from the original source

• Place input video in location videos/Test.mp4.

NOTE: The make_label.py script has boundaries set for this video ~1:20-~1:30. Use this video as Test.mp4 to replicate demo results. Alternatively, refer Further Instructions to set the script up for other videos

• In your python code import Detect() from run.py using

from run import Detect

• Run object detection using

d=Detect(yolo_path="yolo-coco", video_path="videos/Test.mp4", output_path="output/output.avi", confidence=0.5, threshold=0.3)

The above values are set by default and will yield equivalent results even when used as follows

d=Detect()

However, one or more of the default paramters can be changed by explicit mention as shown above.

yolo_path -> Path to directory containing coco.names, yolov3.cfg and yolov3.weights
video_path -> Path to input video
output_path -> Path to output video
confidence -> Minimum probability to filter weak detections
threshold -> Non-maxima suppression threshold i.e. boxes overlapping with a ratio greater than 30% are suppressed for threshold=0.3

Further Instructions

• To use the script for general object detection purposes uncomment lines 82 to 93 and comment lines 93 to 110 in run.py

• To detect other list of items replace list ['sports ball', 'person'] in line 96 of run.py by your desired list of items. Make appropriate changes in if conditional statement at line 103 also.

• To use the colour based team detection for other teams (or videos) replace the dict provided in labeler/boundaries.json by the required values. This dict represent the boundary values within which the jersey colour of a particular team or a category are most likely to be present. The format for this dict is
  {"label1": [[B_lower_limit, G_lower_limit, R_lower_limit], [B_upper_limit, G_upper_limit, R_upper_limit], [B_label_colour, G_label_colour, R_label_colour]], "label2": [[B_lower_limit, G_lower_limit, R_lower_limit], [B_upper_limit, G_upper_limit, R_upper_limit], [B_label_colour, G_label_colour, R_label_colour]], ... }