实时边缘检测：桑德斯视频与图像处理系统研究

需积分: 10 63 浏览量更新于2024-08-02 收藏 1.52MB PDF 举报

本论文标题为《实时边缘检测：使用Sundance视频和图像处理系统》，作者Rajeev Kanwar在2007年至2009年期间，在National Institute of Technology Rourkela的电子与通信工程系攻读硕士学位，专业领域为VLSI设计和嵌入式系统。该研究旨在探讨如何利用Sundance视频和图像处理技术实现实时边缘检测。论文的主要内容涉及前沿的实时（real-time）图像处理技术，特别是针对边缘检测（edge detection），这是一种计算机视觉的基础任务，它能识别出图像中的边界或轮廓，广泛应用于诸如视频监控、自动驾驶、医学成像等领域。Sundance平台作为一个可能的选择，提供了高效且实时的硬件和软件环境，能够支持复杂的图像分析工作。论文可能深入研究了以下关键知识点： 1. **Sundance平台的优势**：可能介绍了Sundance视频和图像处理系统的特性和优势，如高性能硬件加速、低延迟特性，以及与嵌入式系统无缝集成的能力，这对于实现实时边缘检测至关重要。 2. **边缘检测算法**：论文可能会讨论几种经典的边缘检测算法，如Canny边缘检测、 Sobel算子或Prewitt算子，以及如何在Sundance平台上优化这些算法以适应实时性能需求。 3. **系统架构与实现**：可能包含了实时边缘检测系统的整体设计，包括硬件模块（如摄像头接口、图像传感器）、处理单元（如DSP或GPU）、以及软件编程接口和算法执行流程。 4. **性能评估与优化**：论文可能展示了实验结果，包括边缘检测的速度、准确度和功耗等性能指标，以及针对性能瓶颈的优化策略。 5. **应用实例**：论文可能提供了一些实际应用场景，如行人检测、障碍物识别或工业自动化中的应用，来展示实时边缘检测的实际效果和价值。 6. **贡献与局限性**：最后，论文可能会总结研究的贡献，同时讨论存在的挑战和未来改进的方向，以显示对当前技术的理解和对行业的潜在影响。这篇硕士论文探讨的是将Sundance平台与实时边缘检测技术相结合的方法，旨在为VLSI设计和嵌入式系统提供一个高效且实时的解决方案，具有很高的实用性和理论价值。

REAL TIME EDGE DETECTION USING SUNDANCE VIDEO AND IMAGE PROCESSING SYSTEM

National Institute Of Technology Rourkela Page 3

which is not possible in dedicated DSP designs. So FPGAs are ideal choice for

implementation of real time image processing algorithms.

FPGAs have traditionally been configured by hardware engineers using a Hardware Design

Language (HDL). The principal languages being used are VHDL.VHDL are specialized

design techniques that are not immediately accessible to software engineers, who have often

been trained using imperative programming languages. Consequently, over the last few

years there have been several attempts at translating algorithmic oriented programming

languages directly into hardware descriptions. The combination of FPGA AND DSP

processors description language by Sundance module SMT339, allows the designer to focus

more on the specification of an algorithm rather than adopting a structural approach to

coding. For these reasons the used for implementation of image processing algorithms on

FPGA. The SMT339 provides video input and video output using the TMS320DM642 from

Texas Instruments. The hardware is extremely flexible but that comes at the price of

enormous complexity. It can take a long time to determine values for all the relevant

registers of the numerous components in the processor; a slight error can result in bizarre

effects ranging from no input or output at all to strangely corrupted images.

The goal of this thesis is to implement image processing algorithms like convolution and

different edge detectors like a Sobel, Prewitt, Canny and Laplacian edge detection through

Sundance module and compare against the performance of different detector.

Chapter two provides information on the hardware Sundance module SMT339 and prior

related work. Chapter three describes the edge detection algorithms. Chapter four provides

the result and discussion. Chapter five summaries the results and future work.

1.2 LITRATURE REVIEW

In this section, work done in the area of edge detection is reviewed and focus has

been made on detecting the edges of the digital images. Edge detection is a problem of

fundamental importance in image analysis. In typical images, edges characterize object

boundaries and are therefore useful for segmentation, registration, and identification of

objects in a scene. Edge detection of an image reduces significantly the amount of data and

REAL TIME EDGE DETECTION USING SUNDANCE VIDEO AND IMAGE PROCESSING SYSTEM

National Institute Of Technology Rourkela Page 4

filters out information that may be regarded as less relevant, preserving the important

structural properties of an image.

A theory of edge detection is presented. The analysis proceeds in two parts.

(1)Intensity changes, which occur in a natural image over a wide range of scales, are

detected separately at different scales. An appropriate filter for this purpose at a given scale

is found to be the second derivative of a Gaussian, and provided some simple conditions are

satisfied, these primary filters need not be orientation-dependent. Thus, intensity changes at

a given scale are best detected by finding the zero values of for image. The intensity changes

thus discovered in each of the channels are then represented by oriented primitives called

zero-crossing segments. (2) Intensity changes in images arise from surface discontinuities or

from reflectance or illumination boundaries, and these all have the property that they are

spatially localized. Because of this, the zero-crossing segments from the different channels

are not independent, and rules are deduced for combining them into a description of the

image. This description is called the raw primal sketch. The theory explains several basic

psychophysical findings, and the operation of forming oriented zero-crossing segments from

the output of centre-surround filters acting on the image forms the basis for a physiological

model of simple cells.

Due to the large amount of literature on edge detection, I do not intend this to be a

comprehensive literature survey. However, in the following paragraphs, mention some

important works on edge detection and compare results to these methods. Unlike all the

other methods have encountered, the step expansion filter is based on template matching,

and specifically on the novel method of matching using non orthogonal expansion. The step

expansion filter is optimal in the sense of SNR, since Expansion Matching in general is an

optimal SNR template matching method. Other methods follow the approach of defining a

specific set of criteria for the given edge model, and optimizing them, in most cases using

numerical methods. In contrast, the Expansion Matching approach is analytical, and is

generalized so that it easily yields the optimal SNR operator for any desired edge model.

Many works have been performed on obtaining an edge detection operator. Most

noted amongst these is the work of Canny [1] who showed that the ideal operator that

REAL TIME EDGE DETECTION USING SUNDANCE VIDEO AND IMAGE PROCESSING SYSTEM

National Institute Of Technology Rourkela Page 5

maximizes the conventional signal-to-noise ratio in detecting a particular edge is correlation

with the same edge model itself. However, this detection is not well localized and requires

an additional localization criterion. A third criterion that suppresses multiple responses was

also included and numerical optimization resulted in the desired edge detector. Canny‟s edge

detector for step edges is well approximated by the Derivative of Gaussian mask. In

contrast, our edge detector formulation is based on the fact that SNR implicitly defines all

the desired properties of good detection and localization (sharp peak, good localization and

minimal off-center response), and thus given an edge model, the optimal SNR filter for this

model also results in a good edge detector. While Canny [1] worked with finite extent filters.

Deriche [2] used the same approach with infinite extent filters, with the objective of

obtaining an efficient recursive implementation. The resulting operator has the form of an

even, exponentially damped sinusoid and is different from our Step Expansion Filter.

Furthermore, the step expansion filter is also infinite in width, and has an efficient recursive

implementation as well.

Another work using infinite width filters is due to Sarkar and Boyer [3], [4]. In this

work, Canny‟s signal-to-noise ratio and localization criterion, along with another criterion

for spurious response are optimized using the variational approach and nonlinear constrained

optimization. A recursive approximation to these filters is also presented. A comprehensive

set of results with different values of the Multiple Response Criterion (MRC) reveals that for

some values of the MRC, the filters are somewhat similar in appearance to our Step

Expansion Filter. Spacek [5] combined all three of Canny‟s criteria into one performance

measure and simplified the differential equation that yields the optimal filter. To yield the

actual optimal filter, he fixed two of the six parameters involved and determined the

remaining four using boundary conditions. The work of Petrou and Kittler [6] extends

Spacek‟s works for ramp edges.

Shen, Castan, and Zhao [7], [8] present as an optimal operator for edge detection, an

exponential filter. Analytically, this exponential filter is equivalent to the integral of the Step

Expansion Filter that we present. Unlike Shen er al use Expansion Matching and optimize

the SNR criterion, and obtain an exact analytical relationship between the variance of the

expected input noise, and the width (decay parameter in the exponential term) of our Filter.

REAL TIME EDGE DETECTION USING SUNDANCE VIDEO AND IMAGE PROCESSING SYSTEM

National Institute Of Technology Rourkela Page 6

On the other hand, Shen and Castan [7] desire to: a) minimize the energy in the desired

filter‟s response to noise, b) minimize the energy in the derivative of the above noisy

response, and c) maximize the energy of the peak center response to the step edge. Unlike

the SNR, these criteria do not consider the off-center response of the filter to the template (in

this case the step edge model) as undesired noise. Also, their work does not address the

problem of determining an appropriate detector width for a given input noise. Our approach

also offers a general method for easily designing optimal SNR detectors for any edge model,

not only step edges, and can also easily incorporate colored noise models. Another important

point is that the edge maps generated by the two methods are not identical, since we detect

the peaks of the filter output, whereas Shen and Castan obtain the zero crossings of the

output of their exponential filter. The fundamental difference here is that while the step

expansion filter and the exponential filter are related by a simple integral equation, detecting

the peaks of the step expansion filter output in a white noise environment (as per our

analytical model) is not equivalent to detecting the zero crossings of in a white noise

environment (as Shen and Castan propose) since the noise model undergoes a change

(actually becomes more colored and low pass) due to the integration. A more detailed

discussion of this point can be found in the work of Sarkar and Boyer [3].

Modestino and Fries [9] suggest to use the Laplacian of an image in order to detect

edges. In their work they use random fields, and cast the problem as one of obtaining the

minimum mean squared error estimate of the true Laplacian of the actual input image from a

given noisy input image. This obviously results in the Wiener filter of the Laplacian as the

optimal filter. Modestino and Fries do not use this filter itself, but instead use a spatial

frequency weighted version of the Laplacian (basically a Gaussian low-pass spatial filter) to

avoid difficulties in the digital implementation of the optimum Wiener filter. Furthermore,

their work is concentrated on realizing this filter using a recursive implementation. Note that

their Wiener filter has no connection to the SNR optimization that we perform. The Wiener

restoration filter use is based on regarding the given edge model as a blurring function,

which has been shown to be an efficient and regularized implementation of the

nonorthogonal expansion for matching [10].

剩余82页未读，继续阅读

rumon863

粉丝: 0
资源: 6

实时边缘检测：桑德斯视频与图像处理系统研究

Real-Time_Simulator

161204_Mastering_the_FreeRTOS_Real_Time_Kernel-A_Hands-On_Tutorial_Guide

Matlab GUI.rar_For Real_detection_real-time_sleep

Real-Time_High-Resolution_Background_Matting

Real_time_Object_detection_TF-master_objectdetection_源码

CenterMask__Real-Time_Anchor-Free_Instance_Segmen_CenterMask.zip

PoseNet：A Convolutional_networks_for_real-time_6-DOF_camera

Real-Time_3D_Rendering_with_DirectX_and_HLSL

Real-Time_Shader_Programming

Yang_PIXOR_Real-Time_3D_CVPR_2018_paper.zip

最新资源