双目成像系统彩色图像处理流水线融合优化设计方法研究

Binocular camera system has wide application prospects in video monitoring field, virtual reality and mobile terminals fields. How to optimize and design the image processing pipeline of binocular camera has become one of the most pressing problems. In this proposal, we will present a color fusion model for binocular camera by introducing deep learning technique, which based on human vision system and color imaging mechanism. To solve the problem, a robust matching and fusion method for RAW data will be proposed, and the fusion mechanism between the image processing pipeline and advanced vision algorithm will be explored. We plan to provide a solution for the optimal designing of color image processing pipeline of binocular camera to realize the high dynamic and large depth of field video stream output at low optics costs. The advances in the research can provide basic theories and criterions in applications for the development of the fusion design of color image processing pipeline for multi-sensor imaging system. The progresses in the project will promote the popular of binocular camera and improve the ability of chip algorithm design in our country, which have an important scientific value and application value.

双目成像系统在视频监控、虚拟现实与移动终端等领域均具有广泛的应用前景。如何对双目成像系统彩色图像处理流水线进行融合优化设计是双目成像系统应用过程中亟待解决的问题之一。本项目拟通过借鉴人眼视觉成像模型与大脑颜色认知机理，引入深度学习算法构建双目成像系统颜色渲染模型，探索双目成像系统彩色图像处理流水线与高级视觉算法融合优化设计机制，提出基于RAW数据的双目视频流鲁棒性匹配融合方法；研究双目成像系统彩色图像处理流水线融合优化设计理论与方法，以低光学成本实现带有深度数据的高动态大景深视频流实时输出。相关研究的进展和突破不仅有望为多传感器成像设备彩色图像处理流水线融合优化设计提供理论与实践基础，也将推动双目成像设备在视频监控与移动终端等领域的普及，并提高我国图像传感器芯片与处理器芯片算法设计实力，具有重要的科学意义与应用价值。

近年来，多传感器成像技术研究已成为移动终端、视频监控以及工业检测等领域的应用热点，双目成像系统彩色图像处理流水线融合优化设计研究是多传感器成像技术研究的基础。本课题基于人眼成像模型建立双目成像系统RAW数据集，弥补当前双目RAW数据集的空白；研究基于RAW数据的双目/多目高动态彩色图像处理流水线融合设计方法，提出了可抵抗相机抖动的基于RAW数据的多曝光图像融合方法，解决了极端照明环境下基于双目RAW数据的高动态图像捕获问题；研究了场景深度信息获取方法，设计改进的卷积神经网络用于训练双目立体数据；探寻双目成像系统在无人驾驶领域以及农业领域的应用。相关研究的进展和突破可降低双目成像系统相关算法设计复杂度，推动双目成像系统在工业以及农业领域的应用，提高多图像传感器芯片算法设计能力，具有重要的科学意义和应用价值。

内容获取失败，请点击重试