// blob.cpp - HSV segmentation + image moments + depth estimate // Focal length f = 3.22mm; proportionality constant k = 8e-4 cv::Mat hsv; cv::cvtColor(frame, hsv, CV_BGR2HSV); // Threshold for yellow perch marker cv::Mat mask; cv::inRange(hsv, cv::Scalar(20, 50, 100), cv::Scalar(30, 255, 255), mask); // Compute image moments cv::Moments m = cv::moments(mask, true); double cx = m.m10 / m.m00; // centroid x double cy = m.m01 / m.m00; // centroid y // Covariance matrix eigendecomposition → orientation angle double mu20 = m.mu20 / m.m00; double mu02 = m.mu02 / m.m00; double mu11 = m.mu11 / m.m00; double theta = 0.5 * atan2(2 * mu11, mu20 - mu02); // principal axis angle // Depth from focal-length triangulation: Z = f * X_real / X_pixels double blobWidth = sqrt(mu20); // approx pixel width from 2nd moment double Z_est = (8e-4) / blobWidth; // calibrated proportionality posMatrix << cx << " " << cy << " " << theta << " " << Z_est << "\n";

ResearchAI/MLHardware

Monocular Depth Estimation for UAV Perch Landing

Recovering 3D perch pose from one camera without a depth sensor.

0.7 cm

Error @ 25 cm

5.7 cm

Error @ 1 m

3.2 mm

Focal Length

Tech Stack

C++OpenCVHSV SegmentationImage MomentsKalman FilterCMakeVisual Servoing

The problem

A drone landing on a perch has no GPS signal, no laser rangefinder, and no dedicated depth sensor - just a camera. It needs to figure out exactly how far away the landing target is and whether it's correctly aligned, using only what appears in a single video frame. Get the math wrong, and the drone crashes.

The challenge

Autonomous UAV landing on a perch requires knowing the perch's 3D position and the camera's attitude relative to it - without a depth sensor, IMU, or GPS. The system had to recover all of this from a single video frame using only the geometric properties of the detected blob: its centroid, principal axes, and apparent size. The depth estimation formula Z = f·X/X₀ is simple in principle but sensitive to blob detection noise, so the pipeline needed to be robust to partial occlusion and varying lighting.

Architecture & System Design

Single camera estimates 3D target pose using computer vision: image processing extracts target features, geometric math recovers position and orientation from 2D image moments. Enables closed-loop visual control without depth sensor.

Full system schematic available upon request

The pipeline runs in real time on each video frame: (1) Convert BGR → HSV and threshold for the target colour (yellow: H=20-30, S=50-255, V=100-255) to isolate the blob. (2) Compute zeroth, first, and second-order image moments (M00, M10, M01, M20, M02, M11) to recover centroid (cx, cy) and shape covariance. (3) Eigendecompose the 2×2 covariance matrix to get principal axes and orientation angle θ - this gives camera roll/pitch relative to the perch. (4) Depth: Z = f · X_real / X_pixels, where focal length f=3.22 mm was measured via calibration. Output: [cx, cy, θ, Z] written to a position matrix file and used as the visual error signal for the PID controller. A complementary OpenCV toolkit was developed alongside: Canny edge detection with interactive threshold sliders, HSV range calibration tool, and SURF feature matching.

Results

Depth estimation accuracy: 0.67 cm error at 25 cm, 5.7 cm at 1 m, and 25 cm at 2 m - performance degrades linearly with distance as expected from the triangulation model. Orientation estimation successfully recovered camera attitude (roll/pitch) relative to the perch surface at all tested distances. The full pipeline ran at real-time video rates in C++ on a laptop. Results were validated against a Vicon motion capture ground-truth system and documented in the final project report.

Gallery & Demos

Blob Detection Result

HSV segmentation isolating the yellow perch marker with centroid, principal axes, and orientation angle overlaid.

Depth Estimation Geometry

Diagram showing focal-length triangulation: how 2D blob size in pixels maps to 3D distance using calibrated focal length.

Accuracy Performance Plot

Depth estimation error (mm) vs. true distance, showing sub-1cm accuracy within working range.

Click any image or video to expand · ← → keys navigate

More from University of Utah

Vision-Based Autonomous Quadrotor

Researcher

MS thesis project: a quadrotor capable of autonomously taking off, navigating, and perching on branch-like structures using only visual feedback - designed for autonomous crop monitoring in agricultural fields.

OpenCVPID ControlROS

Multi-Arm Coordination - 2-DOF QUANSER

Researcher

Dual-arm robotic manipulation system using 2-DOF QUANSER robots with a master-slave architecture - one arm controlling position, the other controlling force - to collaboratively manipulate objects with precision.

MATLAB/SimulinkQUARCImpedance Control

Adaptive Backstepping - Indoor Micro-Quadrotor

Researcher

Nonlinear controller design for an indoor micro-quadrotor with a suspended pendulum mass - a highly unstable configuration. Adaptive backstepping outperformed classical PID in robustness tests across multiple flight regimes.

MATLAB/SimulinkBackstepping ControlLyapunov Analysis

Sensor-Based SLAM Navigation - iRobot Create

Researcher

Autonomous mapping and navigation system on an iRobot Create platform using IR rangefinders and servo-mounted sensors for 360° SLAM - with RRT path planning to navigate complex maze environments.

MATLABiRobot CreateRRT Path Planning

PUMA 6-DOF Robot Arm - Forward & Inverse Kinematics

Researcher

Forward and inverse kinematics solver for a 6-DOF PUMA 762 robot arm, using Denavit-Hartenberg parameters and an interactive 3D MATLAB GUI with joint sliders, a motion trail, and collision detection.

MATLABDenavit-HartenbergRobotics Toolbox

Sampling-Based & Graph-Search Motion Planning

Researcher

MATLAB implementations of four canonical path-planning algorithms - Dijkstra, A*, PRM, and RRT - applied to a differential-drive robot navigating bitmap maps in configuration space, with real hardware execution on an iRobot Create.

MATLABRRTPRM