本文为荷兰埃因霍温科技大学（作者：Bimpisidis, K.）的硕士论文，共61页。

随着高效算法和低成本高分辨率CMOS相机的出现，图像处理（IP）应用变得越来越流行。然而，IP应用是计算密集型的，消耗大量的能量且处理时间长。最近有人已经提出将图像近似用于这些应用的节能设计，它还减少了长处理时间的影响。这里的挑战是，IP应用程序通常作为更大的闭环控制系统的一部分工作，例如高级驾驶员辅助系统（ADAS）。容忍某些误差对基于图像的控制（IBC）系统的图像近似影响是非常重要的。然而，当IP被近似时，缺乏工具支持来评估这种闭环IBC系统的性能。在这项工作中，我们研究了算法近似对IBC系统控制质量的影响。我们提出了一个闭环的汽车IBC系统的图像近似性能评价框架。我们的框架是用C++编写的，并使用V-REP作为仿真环境。在仿真中，V-REP作为服务器运行，C++模块作为客户端同步模式运行。我们用一个基于视觉的横向控制的例子来说明我们的框架有效性。我们的结果表明，近似计算允许将处理时间提高达3.5倍。在我们的框架上的测量使我们能够对近似影响产生全面的理解，并在使用近似计算时将控制质量提高50%。

Image Processing (IP) applications havebecome popular with the advent of efficient algorithms and low-cost CMOScameras with high resolution. However, IP applications are compute-intensive,consume a lot of energy and have long processing times. Image approximation hasbeen proposed by recent works for an energy-efficient design of theseapplications. It also reduces the impact of long processing times. Thechallenge here is that the IP applications often work as a part of biggerclosed-loop control systems, e.g. advanced driver assistance system (ADAS). Theimpact of image approximations that tolerate certain error on these image-basedcontrol (IBC) systems is very important. However, there is a lack of toolsupport to evaluate the performance of such closed-loop IBC systems when the IPis approximated. In this work, we study the impact of algorithmic approximationon the quality-of-control for IBC systems. We propose a framework forperformance evaluation of image approximation on a closed-loop automotive IBCsystem. Our framework is written in C++ and uses V-REP as the simulationenvironment. For the simulation, V-REP runs as a server and the C++ module as aclient in synchronous mode. We show the effectiveness of our framework using avision-based lateral control example. Our results show that approximatecomputing allows to improve the processing time up to a factor of 3.5. Themeasurements on our framework allowed us to develop a thorough understanding onthe impact of approximation and achieve an overall quality-of-controlimprovement of up to 50%, when using approximate computing.

1. 引言

2. 相关工作

3. 问题描述

4. 基于图像的控制

5. 近似计算

6. 车道偏离检测

7. 实验设置

8. 应用剖析

9. 近似控制下的质量退化

10. 基于控制器设计提高控制质量

11. 结论

更多精彩文章请关注公众号：