Yi Shan

1. Developed Customized Image Restoration Filters with Deep Learning Techniques:

• Engineered a 'Work Filter' using convolutional neural networks (CNNs) for document print quality enhancement, replicating high-end scanner effects.
• Designed a 'Shadow Removal Filter' with advanced image processing algorithms for optimal document texture preservation during reading/documenting processes.
• Implemented an 'Art Filter' using state-of-the-art computer vision techniques for precise scanning of artworks, preserving intricate details and color accuracy.

2. Leveraged Super-Resolution Techniques for Enhanced Old Photo Restoration:

• Employed advanced super-resolution and image restoration techniques to remove stains/creases, and enhance photo resolution, delivering HD electronic photos to users.

3. Boosted User Engagement and Processing Volume with Efficient Algorithm Design:

• Achieved a 2x increase in user base with 450K+ daily users through the application of optimized image processing algorithms, handling the processing of 7M+ images daily.
• Enabled a 5x user growth with high-throughput old photo restoration algorithms, servicing 20K+ daily users and 110K images daily.

1. Applied Image Retrieval Techniques for AGV Navigation:

• Enabled AGVs to quickly locate within navigation maps using image retrieval technology.
• Utilized object detection and tracking technology to identify and track obstacles like pedestrians, vehicles, and shelves, assisting AGVs in obstacle avoidance.
• Achieved a recall and precision of over 99% in real-world environments, significantly enhancing AGV safety.

2. Optimized Image Auto-Exposure Algorithm and Camera Calibration:

• Enhanced image brightness control through image auto-exposure algorithm optimization.
• Developed a binocular camera auto-calibration and verification feature, designed a simplified operation process for users with no prior knowledge to easily complete camera calibration.
• Boosted camera calibration success rate from 67% to 100%, reducing hardware factory repair costs. These improvements resulted in 4 invention patents.

3. Improved BSM Estimation Algorithm for Unmanned Aerial Vehicles (UAVs):

• Optimized binocular disparity estimation (bsm) algorithm to calculate distances of different obstacles in the UAV's direction, enabling automatic obstacle avoidance.
• Applied object detection techniques to identify landing markers, and used pose estimation to calculate UAV's deviation from landing point and feedback to the control system, thus aiding UAVs in automatic landing.