Genetic Algorithms For Parameter Optimization for Disparity Map Generation of Radiata Pine Branch Images

TL;DR

Problem: Safe, efficient UAV branch pruning requires fast, accurate depth estimation from stereo vision; neural methods are accurate but too slow for real-time operation. Approach: A genetic algorithm–based framework automatically tunes nine SGBM and WLS parameters to maximize disparity-map quality while preserving ~0.5 s/frame performance. Contributions: A GA-based optimization framework, a multi-metric evaluation methodology (MSE, PSNR, SSIM), and validation against NeRF-Supervised RAFT-Stereo gold standards on forestry data, with demonstrated generalization across imaging conditions. Findings: GA-optimized configurations achieve a 42.86% reduction in MSE, an 8.47% PSNR increase, and a 28.52% SSIM increase over baseline, with SSIM-optimized parameters offering superior structural fidelity for pruning tasks. Practical impact: Enables safer, more efficient UAV-based forestry pruning under resource constraints, with potential performance gains via implementation optimizations.

Abstract

Traditional stereo matching algorithms like Semi-Global Block Matching (SGBM) with Weighted Least Squares (WLS) filtering offer speed advantages over neural networks for UAV applications, generating disparity maps in approximately 0.5 seconds per frame. However, these algorithms require meticulous parameter tuning. We propose a Genetic Algorithm (GA) based parameter optimization framework that systematically searches for optimal parameter configurations for SGBM and WLS, enabling UAVs to measure distances to tree branches with enhanced precision while maintaining processing efficiency. Our contributions include: (1) a novel GA-based parameter optimization framework that eliminates manual tuning; (2) a comprehensive evaluation methodology using multiple image quality metrics; and (3) a practical solution for resource-constrained UAV systems. Experimental results demonstrate that our GA-optimized approach reduces Mean Squared Error by 42.86% while increasing Peak Signal-to-Noise Ratio and Structural Similarity by 8.47% and 28.52%, respectively, compared with baseline configurations. Furthermore, our approach demonstrates superior generalization performance across varied imaging conditions, which is critcal for real-world forestry applications.

Figures (9)

• Figure 1: Flowchart of traditional disparity map generation process.
• Figure 2: SGBM and WLS processing pipeline for disparity map generation.
• Figure 3: Parameter encoding example for GA optimization of SGBM and WLS parameters, which is a 29-bit integer vector.
• Figure 4: Crossover, mutation, and environment selection in GA.
• Figure 5: Left and right stereo image pairs captured using a ZED Mini camera mounted on a UAV in a Lab environment. Blue markings indicate UAV pruning tools, red markings highlight tree branches, and orange markings identify tree trunks.