Image-Based Visual Servoing UAV

01 Project Overview

Comprehensive evaluation of feature detection algorithms for Image-Based Visual Servoing (IBVS) applications using UAV imagery. Compares SIFT, SURF, ORB, Harris, FAST, and Canny edge detection with Lucas-Kanade optical flow tracking, evaluating performance under realistic conditions for vision-guided robotic control. The system provides the foundation for IBVS control, where visual feedback directly drives robot motion without requiring 3D world models.

02 Key Features & Achievements

Six feature detection algorithms: SIFT, SURF, ORB, Harris Corner, FAST, Canny Edges

Lucas-Kanade optical flow tracking with multi-scale pyramid levels for reliable feature tracking

Comprehensive evaluation metrics: Feature count, processing time, tracking stability, spatial coverage, IBVS performance

Testing framework: Gaussian noise (σ=10, σ=25), blur (5×5, 11×11 kernels), brightness variations (±30)

Real-time visualization: Keypoint overlays, tracking videos, performance charts

UAV dataset processing: 301 frames from UAV123_10fps Boat1 sequence

IBVS performance scoring: Composite metrics combining tracking stability, spatial coverage, control stability, and pose estimation accuracy

03 Technical Stack

Python 3.7+
OpenCV (cv2)
NumPy
Pandas
Matplotlib
Lucas-Kanade Optical Flow
Computer Vision
Image Processing

04 Challenges & Solutions

Evaluating feature detection algorithms for IBVS applications under realistic UAV conditions including noise, blur, and brightness variations

The project required comprehensive evaluation of multiple feature detection algorithms to determine optimal performance for Image-Based Visual Servoing applications. Challenges included maintaining feature stability under adverse conditions, achieving real-time processing speeds, and ensuring robust tracking performance for vision-guided robotic control.

05 Key Achievements

SIFT achieved 88% IBVS performance with 85% tracking stability under adverse conditions

ORB achieved 74.4% IBVS performance with balanced speed and accuracy

SURF achieved 65.5% IBVS performance as faster alternative to SIFT

90% pose estimation accuracy achieved with SIFT algorithm