stephanj / basketballvideoanalysis

Sports Video Analysis

See wiki page for more details.

Court Detection

See details on court detection here.

Mask R-CNN of persons (mask-rcnn)

See details here including online tutorial.

Video output example

https://www.youtube.com/watch?v=yEqGTSd5DQU

Color detection of players (color-detection)

The run script will produce a CSV with the RGB values and percentage of each color.

ToDo: Classify each color result into two groups, identify hopefully each team by color.

Players tracking

See article on players tracking solutions

See also YouTube video with different solutions.

Action Classification

Required so we can classify the game play action of players.

Homography mapping of court

Maps the 3D court onto a 2D version, required for analysis of movement.

See also YouTube video

See details here and article with tutorial.

basketballvideoanalysis's Issues

any screenshots / output?

is the code for auto-encoder available?

HI Thanks you for your amazing work! Question: Is it possible to retrieve and visualize the only the segmented masks of the players (or objects detected) in order to get their jersey colors for exampl?

Cant download "frozen_inference_graph.pb"

Hi, thanks for sharing this project. But I still cant download the "frozen_inference_graph.pb" file. Can you help me ?

Thats great job, how about Football?

Dear Stephan, your project is very interesting, do you have any example for football ??
thank you

Referee detection

You could add referee detection. There are refs in nearly any sport as far as i know. That may helps to distinguish between refs and the away team too, if the model knows that there are 3 different groups.

That also gives us more possibilities in collecting data, since referees in basketball show anything with their hands. It would be nice to learn these gestures(like foul(including kind of foul and the player fouling), 3 point attempt, 3 point made, 2 points made, ft made, travel, double dribbling etc) and translate them into data. With player-detection you can only collect some stats like running distance, field goal attempts. You only know whether a shot counts if you track the ref.

In Addition one camera should be placed behind the game officials to see the refs decisions as good as possible.

Color detection

Current code refers to files in "samples", which are not there.
Also, the run scripts passes in "-i" parameter, but code expects "path" or "p"

(Interesting project!)

We can use action detection technology to detect action.

Have a look at this. It may inspire you.

https://github.com/oulutan/ACAM_Demo

License to use videos

Is there any license to use the videos?

Another company

You could add icebergsports.com to the list, they do hockey analytics from video. I would be interested in helping out and working to have it work with hockey.