MBIE: Motion-Blurred Image Enhancement for Mobile Robots

• Allan Aguiar UFAM
• Laura Martinho UFAM
• João Marcos Cavalcanti UFAM
• José Luiz Pio UFAM
• Felipe Oliveira UFAM

Resumo

Motion blur is a common issue in images captured by mobile robots, often caused by rapid movement or vibrations during navigation. This distortion compromises the robot's ability to accurately perceive and interact with the environment, directly affecting the performance of tasks such as obstacle detection and path planning. In this paper, we propose a deep learning-based method to enhance motion-blurred images, aiming to improve the visual input quality for autonomous navigation. The proposed approach employs a Convolutional Neural Network (CNN) trained to generate sharp images from blurry image inputs. To train and evaluate our model, we used the GoPro dataset, which includes 3214 pairs of blurry and corresponding sharp images. We evaluated the results using standard reference-based quality metrics: Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM). Our method achieved high scores on both quality metrics, outperforming state-of-the-art techniques, indicating improved visual quality and structural consistency in the enhanced images. These results demonstrate that the proposed approach can significantly enhance the perception capabilities of mobile robots operating in dynamic environments, ultimately contributing to safer and more efficient autonomous navigation.

Palavras-chave: Measurement, Deblurring, Visualization, PSNR, Navigation, Distortion, Mobile robots, Convolutional neural networks, Image enhancement, Autonomous robots, Motion Deblurring, Image Enhancement, Au-tonomous Navigation, Mobile Robotics