Links & Resources<\/h2>\n\n- Github Repo:<\/strong> https:\/\/github.com\/abd…<\/a><\/li>\n
- Basketball Detection Dataset:<\/strong> https:\/\/universe.roboflow.com\/…<\/a><\/li>\n
- Zero Shot Classifier:<\/strong> https:\/\/huggingface.co\/…<\/a><\/li>\n
- Basketball Court Keypoint Dataset:<\/strong> https:\/\/universe.roboflow.com\/…<\/a><\/li>\n
- YouTube Video:<\/strong> Build an AI\/ML Basketball Analysis system<\/a><\/li>\n<\/ul>\n<\/div>\n
What if you could analyze a game \u2013 without sensors?<\/h2>\n
Imagine tracking basketball players, analyzing ball possession, calculating team stats and player speed \u2013 all without a single chip or sensor. This AI\/ML project shows how it\u2019s done: with video and vision models only.<\/p>\n
Originally created by Code In a Jiffy, this open-source framework uses YOLO, OpenCV, and Python to detect players, balls, team identities, passes, keypoints, and distances \u2013 in real-time. A showcase of what\u2019s possible with pure AI.<\/p>\n
<\/div>\n![\"Basketball](\"https:\/\/www.xrstager.com\/wp-content\/uploads\/2023\/06\/ai-basketball-analyse-001.jpg\")
<\/p>\n
Real-time YOLO detection of players and refs on the court<\/h6>\n
Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n<\/div>\nZero-shot classification: How AI sees jersey colors<\/h2>\n
To distinguish teams, a zero-shot image classifier was used. Instead of training the model on every jersey combination, prompts like “dark blue t-shirt” guide the AI to assign team colors dynamically.<\/p>\n
This approach adds incredible flexibility. New uniforms? No problem. The classifier adapts based on description, powered by HuggingFace\u2019s NLI models \u2013 reducing prep time and improving scale.<\/p>\n
<\/div>\n![\"Zero-shot](\"https:\/\/www.xrstager.com\/wp-content\/uploads\/2023\/06\/ai-basketball-analyse-002.jpg\")
<\/p>\n
Team detection using zero-shot image classification<\/h6>\n
Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n<\/div>\nUnderstanding the court: Keypoints and layout<\/h2>\n
What happens where? To answer this, the system learns the layout of the court via keypoint detection. A custom-trained model recognizes baskets, corners and zones from any angle.<\/p>\n
This allows all actions to be placed accurately in space. Combined with perspective correction, the data becomes fully spatialized \u2013 a crucial step to build tactical views and performance metrics.<\/p>\n
<\/div>\n![\"Basketball](\"https:\/\/www.xrstager.com\/wp-content\/uploads\/2023\/06\/ai-basketball-analyse-005.jpg\")
<\/p>\n
Landmark-based detection of court positions and orientation<\/h6>\n
Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n<\/div>\nFrom video to top-down tactical map<\/h2>\n
Once court layout is known, the camera view is transformed. Using perspective mapping, the visual field is flattened into a bird\u2019s-eye view \u2013 unlocking powerful analytics.<\/p>\n
Now you can see passes, player movement, team formations and defense gaps \u2013 just like in pro-level dashboards. But this time, it\u2019s AI-generated from video only.<\/p>\n
<\/div>\n![\"AI](\"https:\/\/www.xrstager.com\/wp-content\/uploads\/2023\/06\/ai-basketball-analyse-004.jpg\")
<\/p>\n
Perspective transformation for tactical game analysis<\/h6>\n
Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n<\/div>\nSpeed, distance, performance: In meters, not pixels<\/h2>\n
How fast is each player? How far did they run? The system calculates real-world values by mapping pixel movement to court coordinates \u2013 enabled by AI and geometry.<\/p>\n
This means coaches, analysts and fans can access pro-level stats without any wearables \u2013 just by analyzing the footage with the right models and vision logic.<\/p>\n
<\/div>\n![\"Distance](\"https:\/\/www.xrstager.com\/wp-content\/uploads\/2023\/06\/ai-basketball-analyse-003.jpg\")
<\/p>\n
Movement analysis using real-world coordinates<\/h6>\n
Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n<\/div>\nVideo: AI-powered Basketball Breakdown<\/h2>\n
Watch the full breakdown in motion \u2013 from raw footage to tactical dashboard. The video walks you through every step of the project, showing results in real time.<\/p>\n
If you’re building sports AI, or looking to replace sensors with vision: this is your reference project. Compact, elegant, open-source.<\/p>\n
<\/div>\n
What if you could analyze a game \u2013 without sensors?<\/h2>\n
Imagine tracking basketball players, analyzing ball possession, calculating team stats and player speed \u2013 all without a single chip or sensor. This AI\/ML project shows how it\u2019s done: with video and vision models only.<\/p>\n
Originally created by Code In a Jiffy, this open-source framework uses YOLO, OpenCV, and Python to detect players, balls, team identities, passes, keypoints, and distances \u2013 in real-time. A showcase of what\u2019s possible with pure AI.<\/p>\n
<\/p>\n
Real-time YOLO detection of players and refs on the court<\/h6>\n
Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n To distinguish teams, a zero-shot image classifier was used. Instead of training the model on every jersey combination, prompts like “dark blue t-shirt” guide the AI to assign team colors dynamically.<\/p>\n This approach adds incredible flexibility. New uniforms? No problem. The classifier adapts based on description, powered by HuggingFace\u2019s NLI models \u2013 reducing prep time and improving scale.<\/p>\n Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n What happens where? To answer this, the system learns the layout of the court via keypoint detection. A custom-trained model recognizes baskets, corners and zones from any angle.<\/p>\n This allows all actions to be placed accurately in space. Combined with perspective correction, the data becomes fully spatialized \u2013 a crucial step to build tactical views and performance metrics.<\/p>\n Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n Once court layout is known, the camera view is transformed. Using perspective mapping, the visual field is flattened into a bird\u2019s-eye view \u2013 unlocking powerful analytics.<\/p>\n Now you can see passes, player movement, team formations and defense gaps \u2013 just like in pro-level dashboards. But this time, it\u2019s AI-generated from video only.<\/p>\n Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n How fast is each player? How far did they run? The system calculates real-world values by mapping pixel movement to court coordinates \u2013 enabled by AI and geometry.<\/p>\n This means coaches, analysts and fans can access pro-level stats without any wearables \u2013 just by analyzing the footage with the right models and vision logic.<\/p>\n Image: \u00a9 Code in a Jiffy | Visuals enhanced by Ulrich Buckenlei \u2013 Visoric GmbH<\/sup><\/p>\n Watch the full breakdown in motion \u2013 from raw footage to tactical dashboard. The video walks you through every step of the project, showing results in real time.<\/p>\n If you’re building sports AI, or looking to replace sensors with vision: this is your reference project. Compact, elegant, open-source.<\/p>\nZero-shot classification: How AI sees jersey colors<\/h2>\n
<\/p>\nTeam detection using zero-shot image classification<\/h6>\n
Understanding the court: Keypoints and layout<\/h2>\n
<\/p>\nLandmark-based detection of court positions and orientation<\/h6>\n
From video to top-down tactical map<\/h2>\n
<\/p>\nPerspective transformation for tactical game analysis<\/h6>\n
Speed, distance, performance: In meters, not pixels<\/h2>\n
<\/p>\nMovement analysis using real-world coordinates<\/h6>\n
Video: AI-powered Basketball Breakdown<\/h2>\n