Basketball Shot Analysis

Plotting the trajectory of a basketball shot with Tensorflow's Object Detection API.

Setup:

STEP 1: Clone Tensorflow's Object Detection Repository: https://github.com/tensorflow/models/tree/master/research/object_detection

STEP 2: Follow their installation guide: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md

STEP 3: Download this repository and move all the files into the folder "object_detection" of the tensorflow repository

STEP 4: Place your video of your shot in the same folder and run it.
e.g. How to run it on "test.mp4":

 >>> python3 annotate_video.py test.mp4 

NOTE: If the ball is being shot from the left side of the screen is moving towards the right, add the flag "--left_to_right":

 >>> python3 annotate_video.py test_flipped.mp4  --left_to_right 

STEP 5: Adjust the release point with the slider at the bottom of the window. Pick a frame of the video with the second slider. Close the window.

RESULT: The analyzed picture and the annotated movie have been saved.

Example:

Requirements:

• imageio
  
• numpy
  
• matplotlib
  
• tensorflow
  
• pillow

Tips for best performance:

• Stablize the camera or phone when you are recording the video.
• Cut the video beforehand and feed in short videos to the script.
• Use a high quality video becuase this maximizes the likelyhood of the basketball being recognized.
• Make sure that there is only ONE basketball in the video. Otherwise a "wrong basketball" might be recognized.
• Use a brown basketball.