To visualize and detect the trajectory of a bowling ball from a video

1. Preprocessing the Video
    - Use OpenCV (cv2) to read frames from the video.
    - Convert frames to grayscale and apply filtering to reduce noise.

2. Object Detection (Ball Tracking)
    Multiple techniques to detect the bowling ball:
    A. Background Subtraction
        - If the background is mostly static, use cv2.createBackgroundSubtractorMOG2().
        - This helps to extract moving objects (the ball).
    B. Color-Based Detection (If Ball Has a Distinct Color)
        - Convert the frame to HSV color space.
        - Use cv2.inRange() to create a mask for the ball's color.
    C. Hough Circle Transform (For Spherical Objects)
        - Use cv2.HoughCircles() to detect circular shapes.
    D. Deep Learning-Based Detection
        - If the ball is not easily detected with the above methods, you can use a pre-trained YOLO or SSD object detection model.

3. Tracking the Ball's Trajectory
    Once the ball is detected in each frame:
        - Store its (x, y) coordinates over time.
        - Use matplotlib to visualize the trajectory.

4. Curve Fitting for Smooth Trajectory
    - Use numpy.polyfit() to fit a curve (e.g., quadratic fit) to the detected trajectory.
    - Overlay the predicted trajectory on the video frames.

___________________________________________________________________________________________________________________________________________


| Feature                     | OpenCV                               | CNN                                 |
|-----------------------------|--------------------------------------|-------------------------------------|
| **Speed**                   | ✅ Faster on CPU                     | ❌ Slower without GPU              |
| **Accuracy**                | ❌ Sensitive to noise                | ✅ More robust to variations       |
| **Setup Complexity**        | ✅ No training needed                | ❌ Requires dataset & training     |
| **Lighting Variance**       | ❌ Affected by shadows & reflections | ✅ More resilient                  |
| **Real-Time Performance**   | ✅ Works well on low-power devices   | ✅ If optimized with TensorRT/ONNX |
| **Robustness to Occlusion** | ❌ Fails easily                      | ✅ Handles occlusion better        |
| **Customizability**         | ✅ Easy to tweak parameters          | ❌ Harder to interpret results     |

- If you need a quick, CPU-friendly, and easily interpretable solution:
    OpenCV (edge detection, background subtraction, Hough Circles, or simple tracking).  
- If you need high accuracy and robustness to lighting & occlusions:
    CNN-based object detection (YOLO, SSD, Faster R-CNN).  

___________________________________________________________________________________________________________________________________________

Hybrid Approach (Best of Both Worlds)
1. **Use OpenCV for Fast Preprocessing**  
   - Apply background subtraction or edge detection to isolate regions of interest.  
2. **Use CNN Only for Final Detection**  
   - Instead of running YOLO on the full frame, run it only on extracted regions to speed up processing.  
3. **Tracking with OpenCV (Kalman Filter or Optical Flow)**  
   - Reduce redundant CNN calls by using object tracking after initial detection.  

___________________________________________________________________________________________________________________________________________

Do not stop the recording_2 at frame 200. go ahead