This guide describes how to set up, configure, and run the Vision AI Engine during a live match. The Vision AI system automatically creates ball clips and provides accurate and reliable coding on a variety of supported data detection, giving Video Analysts more time to focus on their analysis of video and data during live matches.
Automatic Vision AI coding currently includes:
- Ball Clip Creation (i.e., Start/Stop Ball Clip on all balls)
- Bowler Action (e.g., Around the Wicket when applicable)
- Keeper Stance (e.g., Keeper Up when applicable)
- Ball Pitch Measurements (i.e., Pitch Map / Line and Length)
- Release Point Measurements
- Bowler Wide on Crease (i.e., Events > Wide on Crease Bowling when applicable)
- Ball Tracking and Pitch Map overlays for use in live streams and highlights packages
- Ball Arrival Measurements (Beta)
- Feet Movement (Beta)
To maximise visibility of the ball, Vision AI must see as much of the pitch as possible above the umpire and bowler. As a general guide, the umpire should not obscure more than one-third of the pitch length. Cameras positioned at a height of 15 metres typically provide sufficient clearance. If this is not possible, we recommend shifting the camera approximately 8 degrees off-centre to reveal more of the pitch.
Vision AI Camera Requirements:
- Stationary cameras at both ends behind the bowler
- Between 0 and 8 degrees off-centre
- 1080p 25FPS or 50FPS video input
- Camera should be zoomed so there is approximately one pitch width either side of the main pitch
- Camera video settings should be optimised for optimal contrast, sharpness, exposure, and shutter speed
For feedback on your camera positioning, vision AI minimum hardware requirements, see: Vision AI Setup Assistant
For post-match Vision AI processing, see: Vision AI Audit
For information about running Vision AI on a separate machine, see: Vision AI Assistant
Vision AI Setup
Before you begin, check the Vision AI hardware requirements page to confirm that your computer meets minimum requirements. If you aren't sure, run the Video > Auto Detect Video Settings tool and apply any recommended changes. Then check the Tools > Configuration > Vision AI tab for confirmation:
After logging in, the latest Vision AI components (approx. 2GB) will automatically begin downloading:
Ensure your device has a stable internet connection, sufficient storage space, and install the update well before your first Vision AI match.
Once downloaded, follow the prompts to install the components and restart NV Play Cricket for Windows.
Vision AI Configuration
On match day:
- Open the Tools > Configuration > Vision AI tab and tick the Enable Vision AI Capture checkbox to activate Vision AI coding.
- Set the Offset from Ball Release duration under Ball Clip Capture. This setting determines how many seconds before release the clip begins (e.g., 2 seconds).
- Set the Clip Duration under Ball Clip Capture. This setting determines total length of the clip (e.g., 10 seconds between the start and end).
In the example below, the chosen settings will generate 10 second ball clips, starting 2 seconds before the bowler releases the ball and finishing 8 seconds after that:
- Select your Detection Settings. These are used to specify the actions that will trigger a ball capture clip. These are enabled by default but can be toggled as needed.
- Select a Live Vision AI Detection Quality option under Quality Settings. This setting specifies the level of computer resources committed to Vision AI coding. Defaults to Medium. Adjust based on computer performance.
- Select a Post Processing Vision AI Detection Quality option under Quality Settings. This setting specifies post-processing quality when Vision AI codes each ball live. Defaults to High. Adjust based on computer performance.
- Select a Second Pass Processing checkbox option. These are:
- Disabled - lowest resource use
- Ball Start Timing Only - verifies and adjusts clip start timing
- Fully Reprocess Ball - recommended for best accuracy and quality.
-
Click OK to save your settings.
Vision AI Capture
Once enabled, a Start Vision AI button appears below the Video Display:
Before play begins, click Start Vision AI to enter Waiting for Capture mode (waiting to detect the bowler entering the frame):
When Vision AI detects the bowler approaching, the software will automatically transition to active mode as if Start Ball Clip had been clicked manually. Vision AI will apply its coding; however, the ball must still be ended manually by the Video Analyst.
Once Second Pass Processing is completed, Vision AI will return to Waiting for Capture, ready for the next ball.
Vision AI coding is displayed in the Ball by Ball panel, and the Vision AI column shows a tick indicating which balls were processed:
- Light Grey = No data detected
- Red = Vision AI Assisted Coding with at least one low-confidence data point
- Black = Vision AI Assisted Coding with all data points of high confidence
If a ball has no Vision AI tick, it means the Start Vision AI mode was not activated in time. You can apply Vision AI retrospectively by right-clicking the ball and selecting Process Vision AI. For bulk processing, go to Video > Vision AI > Process Match Ball Clips, select the required options, then click Start:
Possible Reasons for Missing AI Coding:
- Ball is obscured by the bowler or umpire
- Low video contrast or resolution
- Bowler approach not detected
In such cases, manual edits may be necessary.
Vision AI requires the Vision AI Engine feature licence to be enabled on your NV Play account.