Filtering Based Adaptive Visual Odometry Sensor Framework Robust to Blurred Images.

Area of Science:

• Robotics
• Computer Vision
• Image Processing

Background:

• Visual odometry (VO) accuracy is significantly degraded by blurred images in real-world robotic applications.
• Existing VO methods struggle with motion blur, leading to unreliable pose estimation.
• Developing robust VO for blurred conditions is crucial for practical robot deployment.

Purpose of the Study:

• To present an adaptive visual odometry estimation framework resilient to blurred images.
• To enhance the accuracy and robustness of VO in the presence of image blur.
• To provide a computationally efficient solution for VO in challenging visual conditions.

Main Methods:

• Utilized small image gradient distribution (SIGD) as an objective measure for image blur evaluation.
• Developed an adaptive blurred image classification algorithm to identify and categorize blurred images.
• Proposed an anti-blurred key-frame selection algorithm to maintain VO performance under blur.

Main Results:

• Experimental results demonstrate superior performance of the proposed framework compared to state-of-the-art methods on blurred images.
• The framework effectively mitigates the negative impact of image blur on VO accuracy.
• The approach achieves enhanced robustness without a substantial increase in computational cost.

Conclusions:

• The proposed adaptive framework significantly improves visual odometry robustness and accuracy in blurred image scenarios.
• The SIGD measure and adaptive classification provide effective means to handle image degradation.
• This work offers a practical solution for reliable robotic navigation in visually challenging environments.