Contact person: Fix Stereo Camera Depth Estimation in Python (OpenCV/YOLO pipeline)

Phone:Show

Email:Show

Location: Bowie, United States

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"Description:

I am working on a stereo vision project using Python, OpenCV, and YOLO for object detection. I have already:
• Calibrated my stereo cameras (saved .npz calibration files).
• Built a working pipeline that:
• Captures synchronized video from two Logitech C270 webcams.
• Applies calibration/rectification.
• Runs StereoBM for disparity → depth estimation.
• Runs YOLOv8 object detection on the left camera.
• Displays bounding boxes and estimated depth per object.

The problem:
Despite calibration, my depth estimates are unstable and inaccurate (fluctuating wildly, often reporting “n/a”, or showing wrong distances). For example:
• Objects at 2–3 meters are reported at the wrong distance.
• Disparity maps are noisy (lots of speckles, weak texture).
• Rectification may still have vertical misalignment.

What I need fixed:
• Review my stereo calibration and fix/improve it if necessary.
• Ensure disparity/depth estimation is stable and correct (e.g. ~2 m object reports close to 2 m).
• Ability to tune StereoBM/SGBM parameters or suggest better low-compute alternatives.
• Ensure depth values are output correctly in meters and usable for mapping.
• Deliver clean, commented Python code (building on my existing pipeline, not a major rewrite).

Nice-to-have (but not required):
• Suggestions on integrating depth into a basic mapping pipeline (e.g. point cloud or occupancy grid) or starter code.

Deliverables:
• Fixed/calibrated Python script (starting from my existing code).
• Documentation/comments so I can understand what changed.
• Verification example: Demonsteate with your webcam stable depth values for detected objects at known distances.

Skills needed: Python, OpenCV (StereoBM/SGBM), stereo calibration, computer vision." (client-provided description)