Visual Servoing Method Based on Partial Convolution Acceleration and Improved Kalman Filter for Robotic Arm

To address slow target recognition speed, low accuracy, and the limited robustness of traditional Kalman filtering under non-Gaussian noise in robotic arm visual servo control, a method combining partial convolution-based rotational bounding box detection and optimized Kalman filtering with the regularized maximum correntropy criterion is proposed. The method enhances feature point recognition efficiency by integrating a partial convolution module into the YOLOv11-OBB network and improves Kalman filtering robustness to nonGaussian noise. A particle swarm optimization algorithm is used to tune the Gaussian kernel bandwidth. Experimental results show that the method increases recognition speed to 37.6 FPS, improves smoothness by $25.48 \%$, reduces error by $62.85 \%$, and boosts grasping success rates by 4.65%, 9.72%, and 12.70% under varying interference levels.