Skip to content

Computer Vision: Real‑Time Stereo Depth Estimation

Depth pipeline using StereoSGBM/SBM with calibration & rectification

Role: Solo developerTimeline: May-August 2025
OpenCVPythonCamera Calibration3D ReconstructionReal‑time

Problem → Approach

  • Needed accurate, real‑time depth for robotics navigation on low‑cost sensors.
  • Built full pipeline: intrinsic/extrinsic calibration, rectification, SGBM disparity, post‑filtering.
  • Designed 3D‑printed stereo mount to ensure pixel‑level horizontal alignment.

Key Engineering

  • Checkerboard calibration & lens distortion correction.
  • Epipolar rectification and disparity benchmarking (Middlebury dataset).
  • Planned Kalman filtering for temporal noise suppression.

Impact

  • Reliable depth maps suitable for embedded navigation & obstacle avoidance.
  • Validated with public datasets and live camera feeds.
Source on GitHub Live demo Processing video
Tsukuba stereo input pair Ground truth disparity Left camera feed Right camera feed

Autonomous Drone Navigation System

Custom flight‑controller hardware + CV stack for autonomous flight

Role: Systems & firmwareStatus: In flight‑tuning
ESP32freeRTOSRaspberry PiSLAMPID ControlKiCAD

Architecture

  • Custom PCB with power, sensor, and motor control; prioritized IMU pipeline on freeRTOS.
  • Pi handles vision: stereo depth & object detection feeding obstacle avoidance.
  • Multi‑loop PID for attitude/altitude; A*/RRT for path planning.

Hardware & Software

  • MPU‑9250 IMU, GPS, barometer; Kalman‑based fusion and vibration mitigation.
  • SLAM for localization; mission execution via waypoint planner.

Current Results

  • Hardware complete; PID tuning in progress for first autonomous missions.
  • Vision stack integrated; next phase is flight envelope characterization.
Custom flight controller PCB Final custom drone build
Short demo

Garage Parking Assistant

Ultrasonic‑guided parking with real‑time distance feedback

Role: Embedded developer
ArduinoUltrasonic SensorsEmbedded

Implementation

  • Sensor polling with debounced readings; smoothing filter for stable output.
  • Simple UI cues for driver attention and safety.

Impact

  • Improved parking consistency; prevents front‑bumper contact in tight spaces.

Contact

Open to internships/co-ops in robotics, embedded systems, and computer vision. Email me • GitHub • Home