Score2Stream is a free, open-source tool that reads scoreboards of sports events and other digital displays using a video camera, recognizes the displayed values through image processing, and outputs the result as a TV graphic for live streams. The graphic is served via a built-in web server as an HTML page and can be integrated directly into tools like OBS or other streaming software.
The idea was inspired by the project Scoreboard-webcam-OCR by Xiaoyang Kao.
No training required: Score2Stream ships with pre-built sample data for a wide range of common digital sports displays. Once a segment is placed on the scoreboard, the digits are typically recognized immediately — out of the box, no manual training needed. Custom samples can be added if required, but are usually unnecessary for standard displays.
The tool consists of three tabs: Input, Templates, and Board.
The Input tab is where the video source is configured. Clicking the Inputs button lists all currently available cameras or video files. The list refreshes on each click. Once a source is selected, it is displayed in the video view on the left.
The image can be repositioned (left/right, up/down) and rotated using the controls in the Video handling section. These adjustments are preserved after segments are defined.
Segments are rectangular regions drawn directly onto the video image, each covering one display area of the scoreboard — for example, a single digit of the game clock or the score. Each segment is assigned a meaning via the Type dropdown, such as Game mins, Game secs, Shot clock, Score home, Score guest, Period, and others.
Additional processing options:
| Option | Description |
|---|---|
| Image queue size | Number of consecutive frames merged into a single output image before processing. Useful for LED displays whose refresh rate exceeds the camera frame rate, causing individual frames to appear incomplete. |
| Processing delay [ms] | Delay between processing steps in milliseconds. Primarily useful when testing with imported video files that would otherwise play back too fast. |
| Confidence at [%] | Minimum recognition confidence in percent. If the match falls below this threshold, the Empty value defined in the Templates tab is output instead. |
| Wait to match [ms] | Time in milliseconds a changed value must persist before it is accepted. Prevents flickering caused by briefly unstable input. |
The Templates tab is used to define reference images (samples) for recognition. A template is assigned to one or more segments and contains the sample images against which the live camera frames are compared.
Samples can be added in two ways:
- Manually: Select a segment, wait for the desired value to appear, and click Add as sample to capture it as a reference image. Then enter the correct value in the text field.
- Automatically (Detect samples): The tool automatically captures new samples whenever the similarity to all existing samples falls below the threshold set in New sample at [%]. For example, at 90%, a new sample is created as soon as the current image differs by more than 10% from all known samples.
Additional settings:
| Option | Description |
|---|---|
| Max nr undetects | Maximum number of unrecognized frames before a new sample is created automatically. |
| Empty value | Value output when recognition confidence falls below the Confidence threshold. Can be left blank or set to a placeholder such as -. |
Note: Deactivate the detection mode once enough samples have been collected.
The Board tab controls the graphic output. Recognized values are transmitted to the HTML graphic page via a built-in web server and a WebSocket connection as a JSON stream.
On the right-hand side, values can be entered manually or used to override automatically detected values by enabling No detection. Changes are not transmitted to the graphic immediately — the Update board button must be clicked to apply them. The button activates automatically as soon as any value on the right-hand panel is modified.
Available actions and settings:
| Option / Action | Description |
|---|---|
| Update board | Applies changed values to the graphic output. |
| Open board | Opens the current scoreboard graphic in the local browser. |
| Restart server | Restarts the built-in web server. |
| Open root folder | Opens the server folder where custom HTML graphic pages can be placed. |
| Port Server / Port Socket | Ports used for the HTTP server and the WebSocket connection. |
| Delay Socket [ms] | Update interval of the WebSocket data transmission in milliseconds. |
In addition to the included sample scoreboard, custom HTML graphics can be created and placed in the root folder. The graphic page receives all values — score, game clock, shot clock, team names, period, fouls, and more — via WebSocket as a JSON stream and can render them in any desired layout.
The output is designed to be used as a browser source in OBS or similar streaming tools, with full support for transparency (chroma key or CSS-based).
| Setting | Description |
|---|---|
| Image queue size | Frames merged per output image (reduces LED flicker artifacts). |
| Processing delay [ms] | Delay between capture cycles (useful for video file testing). |
| Confidence at [%] | Minimum match confidence; below this, the empty value is shown. |
| Wait to match [ms] | Minimum duration a value must be stable before being accepted. |
| New sample at [%] | Similarity threshold below which a new sample is auto-created in detection mode. |
| Max nr undetects | Cap on unrecognized frames before auto-sample creation triggers. |
| Empty value | Fallback value shown when no confident match is found. |
| Delay Socket [ms] | WebSocket update interval for the graphic output. |