Advanced Python Script for AI Carrom Game
Budget: ₹600 – ₹1,500 INR
Advanced Python Script for Carrom AI Assistant — Screen Capture, Object Detection, and Trajectory Prediction from Video
? Project Description:
I need a powerful and memory-optimized Python script designed to work as an AI Carrom Game Assistant. Here's the complete workflow the script should follow:
? Functionality Requirements:
Video Input:
I will upload a Carrom board gameplay screen recording (from my phone) to my desktop.
The script will work only on screen-captured gameplay, not directly on video files.
Screen Capture:
The script will periodically capture screenshots from the desktop (where the video is playing).
Before capturing the screenshot, the user should click Enter within 5 seconds to activate ROI (Region of Interest) selection using a rectangle draw.
This selected ROI becomes the processing region (to avoid mirror effects and speed up detection).
Object Detection:
My trained model (built via Roboflow, format available) already detects:
Striker
White token
Black token
Pocket
Yellow aiming line
Boundary
The script must detect objects from each screenshot using this model (ONNX or YOLO format).
Calibration from Video:
Object size calibration must be automated from the video itself:
Striker radius
Token radius
Velocity estimation (from movement in frames)
Aiming line detection
The script must start trajectory prediction only after the yellow aiming line is detected.
Physics & Trajectory Engine:
After detection, the script should:
Predict possible striker paths based on angle and power.
Simulate and draw all possible collisions:
Token-to-token
Token-to-boundary
Striker-to-token
Striker-to-boundary
Handle realistic physics: reflections, momentum transfer, and friction.
Live Drawing:
The script should draw the predicted trajectories directly on the screenshot using colored lines.
Result should be shown as a video or screen recording, not just still images.
Performance & Memory:
Existing script (which I’ll provide) works but has memory leaks and random crashes (processing window terminates).
Fix these memory/optimization issues for stable long-term usage.
Model Integration:
Sometimes, the model gives incorrect predictions — even though it’s trained on 4K+ images.
Need to verify model output consistency and improve fallback handling (retry detection or confidence threshold adjustment).
? What I Will Provide:
My partially working codebase (trajectory + detection working).
My trained object detection model (ONNX or YOLO format).
Sample screen recording videos.
Sample screenshots for calibration logic reference.
? Expected Deliverables:
✅ Final Python script with:
Optimized screen capturing
ROI selector
Detection + Trajectory prediction
Video overlay of all predicted physics
✅ Full working video output showing the entire flow
✅ Clean and modular code (with comments)
✅ Debugged memory usage and performance
? Tip for Developers:
If you’ve heard of or seen "Bitaim AI Carrom Assistant", this project is quite similar — but made from scratch using custom models and physics.
? Notes:
Please bid only if you have experience with:
OpenCV
YOLO/ONNX object detection
Physics simulation in 2D
'See screenshots (562,563,564) i need like this', 'current result 555,58,59'
Real-time image processing & optimization
? Project Description:
I need a powerful and memory-optimized Python script designed to work as an AI Carrom Game Assistant. Here's the complete workflow the script should follow:
? Functionality Requirements:
Video Input:
I will upload a Carrom board gameplay screen recording (from my phone) to my desktop.
The script will work only on screen-captured gameplay, not directly on video files.
Screen Capture:
The script will periodically capture screenshots from the desktop (where the video is playing).
Before capturing the screenshot, the user should click Enter within 5 seconds to activate ROI (Region of Interest) selection using a rectangle draw.
This selected ROI becomes the processing region (to avoid mirror effects and speed up detection).
Object Detection:
My trained model (built via Roboflow, format available) already detects:
Striker
White token
Black token
Yellow aiming line
Boundary
The script must detect objects from each screenshot using this model (ONNX or YOLO format).
Calibration from Video:
Object size calibration must be automated from the video itself:
Striker radius
Token radius
Velocity estimation (from movement in frames)
Aiming line detection
The script must start trajectory prediction only after the yellow aiming line is detected.
Physics & Trajectory Engine:
After detection, the script should:
Predict possible striker paths based on angle and power.
Simulate and draw all possible collisions:
Token-to-token
Token-to-boundary
Striker-to-token
Striker-to-boundary
Handle realistic physics: reflections, momentum transfer, and friction.
Live Drawing:
The script should draw the predicted trajectories directly on the screenshot using colored lines.
Result should be shown as a video or screen recording, not just still images.
Performance & Memory:
Existing script (which I’ll provide) works but has memory leaks and random crashes (processing window terminates).
Fix these memory/optimization issues for stable long-term usage.
Model Integration:
Sometimes, the model gives incorrect predictions — even though it’s trained on 4K+ images.
Need to verify model output consistency and improve fallback handling (retry detection or confidence threshold adjustment).
? What I Will Provide:
My partially working codebase (trajectory + detection working).
My trained object detection model (ONNX or YOLO format).
Sample screen recording videos.
Sample screenshots for calibration logic reference.
? Expected Deliverables:
✅ Final Python script with:
Optimized screen capturing
ROI selector
Detection + Trajectory prediction
Video overlay of all predicted physics
✅ Full working video output showing the entire flow
✅ Clean and modular code (with comments)
✅ Debugged memory usage and performance
? Tip for Developers:
If you’ve heard of or seen "Bitaim AI Carrom Assistant", this project is quite similar — but made from scratch using custom models and physics.
? Notes:
Please bid only if you have experience with:
OpenCV
YOLO/ONNX object detection
Physics simulation in 2D
'See screenshots (562,563,564) i need like this', 'current result 555,58,59'
Real-time image processing & optimization