CapnaDSP: SPECTER - System-level Programming Environment and Common Framework for Intelligent Image Processing

近年来，传感器技术的进步与高性能嵌入式处理器的结合产生了许多新颖的应用和产品，例如自动泊车汽车、血管介入的实时可视化、车辆跟踪和3D图像重建。这些设计中的限制因素通常不在于设计者的创造性，而在于设备的底层性能和对其进行编程所需的努力。通过提供一个开放、强大、该项目针对不断增长的智能成像市场，该市场的应用程序要求不断提高性能，并转向更高的分辨率、帧速率和多源成像技术，以提高交付给用户的内容价值。在安全方面，下一代CCTV不再依赖于操作员驱动的独立摄像机平台，而是依赖于上下文和位置感知辅助操作，其中各个图像馈送被组合以建立增强的场景感知，例如，当感兴趣区域在一个摄像机馈送中扩展时;这需要被传送以允许多个图像的动态拼接，此外，还需要采用其他技术，例如将来自射频频谱（可见光和红外线）不同部分的图像源结合起来。例如，当可见光减少时，合并红外图像以提高质量。由于本地数据存储成本高昂，并且无法分发数据量，因此需要对数据的持续时间和保留做出明智的决策。虽然这些技术在本质上是复杂的，但它们依赖于一组基本的操作块来促进它们的实现，如果在模块化的、可扩展的硬件加速处理平台中提供这些操作块，则这些操作块将被证明是优于其他选项的主要卖点。FPGA硬件加速已应用于高端系统，但广泛的市场采用将需要降低进入中端和小批量细分市场的门槛。正是在提供一个通用的模块化解决方案，可以快速定制和实施，CapnaDSP看到了一个强大的和迅速崛起的市场竞争力。我们相信，通过最终使这项技术通过一个易于配置的即插即用编程工具的COTS平台，我们可以把一个产品推向市场，具有真正的独特卖点和商业吸引力。这将打开FPGA硬件加速以前不可行的细分市场。从300万像素的CCD设备到800万像素以上的CMOS相机，帧速率从10- 30 fps上升到60 fps以上，这意味着在很短的时间内，数据量将要求新的处理方法。正是在这一点上，证明了我们的方法在这个项目中的优点，我们将寻求商业化的输出，并介绍我们的即插即用处理解决方案，以一系列的市场。

Advances in sensor technology combined with high performance embedded processors hasyielded many novel applications and products in recent years, such as self-parking cars, realtimevisualisation for vascular intervention, vehicle tracking, and 3D image reconstruction.The limiting factor in these designs is generally not in designer inventiveness but in theunderlying performance of the device and effort required to program it. By providing anopen, powerful, and easy to program platform, we address a number of key markets.The project is targeted at the growing intelligent imaging market where applications aredemanding ever increasing performance with the shift towards higher resolutions, frame ratesand multi-source imaging techniques to improve content value delivered to users. In security,next generation CCTV can no longer rely on operator-driven, standalone camera platforms buton context and position-aware assisted operation where individual image feeds are combinedto build up an increased scene awareness, e.g. as a region of interest is expanded in onecamera feed; this needs to be communicated to allow dynamic stitching of multi-sourceimages to maintain full scene coverage.Further techniques such as combining image sources from different parts of the radiofrequency spectrum, visible and infra-red, are also needed. For example, merging infra-redimages to increase the quality as the visible light decreases. As local data storage is expensiveand distributing data volumes is prohibitive, smart decisions on duration and retaining of datais required. While complex in their nature, these techniques rely on a set fundamentaloperational blocks to facilitate their implementation which if provided in a modular,extensible hardware accelerated processing platform, would prove a major selling point overother options. FPGA hardware acceleration has been applied to high end systems butwidespread market adoption will require a lowering of the barriers to entry for mid-range andlow-volume market segments. It is in providing a generic modular solution that can be quicklycustomised and implemented that CapnaDSP sees a strong and rapidly emerging marketopportunity.We believe that by ultimately making this technology accessible through a COTS platformwith easily configurable plug and play programming tools we can bring a product to marketthat has a genuine unique selling point and commercial appeal. This will open up marketsegments where FPGA hardware acceleration was not previously viable. Moves from 3 megapixelCCD devices to CMOS cameras in the 8+ mega-pixel range, with frame rates risingfrom 10-30fps to 60fps+, will mean that in a very short space of time the amount of data willmandate new processing approaches. It is at this point that having demonstrated the merit ofour approach in this project, we will seek to commercialize the output and introduce our plugand play processing solution to a range of markets.