SURECAVI - Super-resolution for visible camera system

Abstract

Centre Borelli's recent advances in super-resolution for remote sensing applications demonstrate the maturity of themse image and video restoration techniques, emphasizing their significance in commercial and defense uses. LERITY specializes in high-definition vision systems, particularly in degraded visibility conditions. They actively pursue research and innovation in advanced signal processing algorithms, optronic design, and sensor integration. Among their products, the new algorithms improve systems like CATEYE System and CATEYE System XLR. However, due to computational constraints, these algorithms are not yet planned for embedded architectures.

The project aims to enhance camera resolution and image quality through software and hardware improvements, while maintaining a similar price point, achieving at least a 1.5x increase in range. Super-resolution techniques will merge multiple raw images to produce a super-resolved image, enhanced through real-time deblurring and denoising. These operations require increased computing power. Fortunately, advancements in GPUs and parallel algorithms allow for real-time processing. Centre Borelli possesses a range of algorithms, adapting the processing chain to computational constraints. Improved quality will enhance detection, recognition, and reduce false alarms.

This project details the necessary hardware and algorithmic adaptations for CATEYE System cameras. Hardware optimizations include optics’ PSF, sensor micro displacements, bandwidth optimization, and integration of a wide field color camera. Software adaptations cover deblurring, recalibration in noise, image fusion, denoising, saturated pixel management, and motion detection for artifact-free fusion. Optical imaging poses specific challenges such as increased noise, color matting, vignetting, and uncompensated motion or vibration in the scene. The project aims to develop a prototype of a super-resolved visible observation system and real-time processing chain, achieving a TRL of 6.

Partners

Members

Publications