Skip to content

Latest commit

 

History

History
30 lines (18 loc) · 889 Bytes

README.md

File metadata and controls

30 lines (18 loc) · 889 Bytes
Raw

BallDetection

Implementation of Machine learning algorithms to detect a goal in basketball games.

  • Naive classifier
  • Logistic regression
  • SVM
  • Multiple layer perceptron
  • CNN

Introduction

Dataset

Our dataset is four basketball videos containing about 400, 000 frames.

Annotation

Basically, we label each frame as goal(1) or not goal(0). A frame is labeled as goal if from the time when the bottom of the basketball touches the loop to the time when the top of the ball is leaving the nets.

Preprocess

Label, cut, and randomly shuffle the frames. See details in preprocess.

Feature extraction

For naive classifier, LR, SVM, we use HoG to extract features.

Evaluation

A ROC curve is used for evaluation.