전방위 이미지의 파노라마 영상처리를 위한 저비용 하드웨어 구조에 관한 연구

Abstract

전방위 이미징 (catadioptric omnidirectional imaging) 시스템은 포물선 모양 또는 원뿔 모양의 거울과 렌즈를 사용하여 시야각이 360도인 이미지를 생성한다. 생성된 이미지는 중심이 같은 원형의 영상 왜곡이 발생하기 때문에 직사각형의 파노라마 이미지로 변환되어야 한다. 많은 응용사례에서 사용되는 전방위 이미징 (catadioptric omnidirectional imaging) 시스템은 특히 고해상도의 이미지에서 실시간, 저비용으로 영상처리를 해야한다. 본 논문에서는 소프트웨어와 하드웨어를 적절히 결합하여 저비용 하드웨어 구조를 제안하였고, 몇 가지 최적화 전략을 사용하였다. 좌표변환 알고리즘에 의해 발생되는 직접적인 나눗셈 연산을 피하기 위하여 소프트웨어로 좌표값을 미리 계산하여 룩업테이블 (look-up table)에 저장하였고, 전방위 이미지 (catadioptric omnidirectional image)의 대칭 특성을 이용하여 계산량을 줄이고 룩업테이블(look-up table)의 사이즈를 줄였다. 이미지 좌표 변환 시 발생되는 손실된 화질을 개선하였고, 최근방 이웃 보간법 (nearest neighborhood interpolation)보다 이미지 화질은 높이고 양선형 보간법 (bilinear interpolation)보다 계산량을 줄이기 위하여 9-point 보간법을 사용하였다. 실험 결과 제안한 하드웨어 구조는 기존의 방법보다 룩업테이블 (look-up table)의 사이즈를 약 68% 줄였고, 최대 87.5%까지 줄일 수 있다.|Catadioptric omnidirectional imaging systems generate images of 360 degrees field of view using lense and mirrors such as parabolic mirror or hyperboloid mirror. The omnidirectional images need to be transformed into rectangular panorama images, because of the concentric circular distortion. In most application scenarios, it has to perform the panorama unrolling in real-time and at low-cost, especially for high-resolution images. In this paper, a low-cost hardware design is proposed, which adopts a software/hardware cooperative structure and employs several optimization schemes. To avoid the on-line division operation caused by the coordinate transformation algorithm, it is pre-computed and stored in the form of a look-up-table (LUT). And then, the amount of memory used by the LUT is reduced by using symmetrical characteristic. The 9-point interpolation is used to avoid complex computation and to improve the output image quality. Experimental results show that the proposed hardware structure reduced the size of LUTs by 68%, and it can reduce by maximum 87.5%. Also, proposed hardware architecture achieves an effective real-time performance and lower implementation cost, and it can be applied to other kinds of catadioptric omnidirectional imaging systems.; Catadioptric omnidirectional imaging systems generate images of 360 degrees field of view using lense and mirrors such as parabolic mirror or hyperboloid mirror. The omnidirectional images need to be transformed into rectangular panorama images, because of the concentric circular distortion. In most application scenarios, it has to perform the panorama unrolling in real-time and at low-cost, especially for high-resolution images. In this paper, a low-cost hardware design is proposed, which adopts a software/hardware cooperative structure and employs several optimization schemes. To avoid the on-line division operation caused by the coordinate transformation algorithm, it is pre-computed and stored in the form of a look-up-table (LUT). And then, the amount of memory used by the LUT is reduced by using symmetrical characteristic. The 9-point interpolation is used to avoid complex computation and to improve the output image quality. Experimental results show that the proposed hardware structure reduced the size of LUTs by 68%, and it can reduce by maximum 87.5%. Also, proposed hardware architecture achieves an effective real-time performance and lower implementation cost, and it can be applied to other kinds of catadioptric omnidirectional imaging systems.