Fast training for object recognition with structure-from-motion

In this paper we present a system for statistical object classification and localization that applies a simplified image acquisition process for the learning phase. Instead of using complex setups to take training images in known poses, which is very time-consuming and not possible for some objects, we use a handheld camera. The pose parameters of objects in all training frames that are necessary for creating the object models are determined using a structure-from-motion algorithm. The local feature vectors we use are derived from wavelet multiresolution analysis. We model the object area as a function of 3D transformations and introduce a background model. Experiments made on a real data set taken with a handheld camera with more than 2500 images show that it is possible to obtain good classification and localization rates using this fast image acquisition method. The text was submitted by the authors in English. Marcin Grzegorzek, born in 1977, obtained his Master’s Degree in Engineering from the Silesian University of Technology Gliwice (Poland) in 2002. Since December 2002 he has been a PhD candidate and member of the research staff of the Chair for Pattern Recognition at the University of Erlangen-Nuremberg, Germany. His fields are 3D object recognition, statistical modeling, and computer vision. He is an author or coauthor of seven publications. Michael Reinhold, born in 1969, obtained his degree in Electrical Engineering from RWTH Aachen University, Germany, in 1998. Later, he received a Doctor of Engineering from the University of Erlangen-Nuremberg, Germany, in 2003. His research interests are statistical modeling, object recognition, and computer vision. He is currently a development engineer at Rohde & Schwarz in Munich, Germany, where he works in the Center of Competence for Digital Signal Processing. He is an author or coauthor of 11 publications. Ingo Scholz, born in 1975, graduated in computer science at the University of Erlangen-Nuremberg, Germany, in 2000 with a degree in Engineering. Since 2001 he has been working as a research staff member at the Institute for Pattern Recognition of the University of Erlangen-Nuremberg. His main research focuses on the reconstruction of light field models, camera calibration techniques, and structure from motion. He is an author or coauthor of ten publications and member of the German Gesellschaft für Informatik (GI). Heinrich Niemann obtained his Electrical Engineering degree and Doctor of Engineering degree from Hannover Technical University, Germany. He worked with the Fraunhofer Institut für Informationsverarbeitung in Technik und Biologie, Karlsruhe, and with the Fachhochschule Giessen in the Department of Electrical Engineering. Since 1975 he has been a professor of computer science at the University of Erlangen-Nuremberg, where he was dean of the engineering faculty of the university from 1979 to 1981. From 1988 to 2000, he was head of the Knowledge Processing research group at the Bavarian Research Institute for Knowledge-Based Systems (FORWISS). Since 1998, he has been a spokesman for a “special research area” with the name of “Model-Based Analysis and Visualization of Complex Scenes and Sensor Data” funded by the German Research Foundation. His fields of research are speech and image understanding and the application of artificial intelligence techniques in these areas. He is on the editorial board of Signal Processing, Pattern Recognition Letters, Pattern Recognition and Image Analysis, and the Journal of Computing and Information Technology. He is an author or coauthor of seven books and about 400 journal and conference contributions, as well as editor or coeditor of 24 proceedings volumes and special issues. He is a member of DAGM, ISCA, EURASIP, GI, IEEE, and VDE and an IAPR fellow.     

Fast Phong明暗处理算法的FPGA实现

Phong明暗处理算法是生成真实感3D图像的最佳算法之一。VLSI技术的发展以及对于高真实感实时图形的需求使得Phong明暗处理算法的实现成为可能。利用泰勒级数近似的Fast Phong明暗处理算法适合硬件实现。但是用硬件实现此算法需要存储大量数据的ROM表,这增加了实现的难度。通过误差分析,提出了优化的查找表结构。通过在FPGA上对所提结构进行验证,结果表明,该方案在提高速度、精度的同时减小了硬件资源的使用量。     

Computer vision-based object recognition for the visually impaired in an indoors environment: a survey

Though several electronic assistive devices have been developed for the visually impaired in the past few decades, however, relatively few solutions have been devised to aid them in recognizing generic objects in their environment, particularly indoors. Nevertheless, research in this area is gaining momentum. Among the various technologies being utilized for this purpose, computer vision based solutions are emerging as one of the most promising options mainly due to their affordability and accessibility. This paper provides an overview of the various technologies that have been developed in recent years to assist the visually impaired in recognizing generic objects in an indoors environment with a focus on approaches based on computer vision. It aims to introduce researchers to the latest trends in this area as well as to serve as a resource for developers who wish to incorporate such solutions into their own work.     

基于FPGA的相关干涉仪测向算法的设计与实现

为了适应日渐复杂的电磁对抗环境,有别于传统的处理器采用串行架构使用IQ数据进行测向算法运算,为了能更快速、更稳定的计算出相位差,提出一种优化后的相关干涉仪测向算法架构并在FPGA上实现设计。详细给出FPGA测向系统架构和运算流程,并结合实际采取一个稳定的、经济实惠的系统架构。为了满足后续实际应用需求,在Xilinx的xc7vx690tffg1927-2为逻辑控制单元的载板进行板级测试中,验证了FPGA进行相关干涉仪测向识别结果的稳定性和高效性,经过验证发现,相较于传统的FPGA相关干涉仪测向架构,经实验测试后实现了所提出的架构在LUT资源占用率上降低了39.5%,测向速度提高17%,可处理信号带宽80MHz,跳速高达1200跳/s。     

Extraction of spatial-temporal features for vision-based gesture recognition

One of the key problems in a vision-based gesture recognition system is the extraction of spatial-temporal features of gesturing.In this paper an approach of motion-based segmentation is proposed to realize this task.The direct method cooperated with the robust M-estimator to estimate the affine parameters of gesturing motion is used.and based on the dominant motion model the gesturing region is extracted,i.e.,the dominant object.So the spatial-temporal features of gestrues can be extracted.Finally,the dynamic time warping(DTW) method is directly used to perform matching of 12 control gestures(6 for “translation“ orders,6 for “rotation“orders).A small demonstration system has been set up to verify the method,in which a panorama image viewer can be controlled(set by mosaicing a sequence of standard“Garden“ images)with recognized gestures instead of the 3-D mouse tool.     

Grasping the not-so-obvious: vision-based object handling for industrial applications

This article reports on several industrial applications of a visually guided system for robot grasping using an inexpensive two-finger gripper. In all cases, the robot uses visual information as input and is able to reason about the shapes of objects in a scene in order to decide the best stable grasp online. The first version of this system was able to grasp rectangular parts in arbitrary positions in the scene and was successfully deployed. New applications in industry have been addressed that must cope with the cost, time, and reliability requirements imposed by the industrial process. The results show that the capabilities of the underlying methodology make it feasible to deal with more complex shapes, even a priori unknown, opening up new possibilities within industrial domains, such as the food industry, that has traditionally not been fully automated due to a large variability in the shaper of the objects to be handled.     

说话人识别系统中VQ判决模块的FPGA实现

目前,基于PC或DSP的系统其处理能力无法满足海量语音信号高速处理需求的增长。本文分析了VQ(矢量量化)搜索算法的硬件实现复杂度,针对说话人识别过程中运算量最大、耗时最长的判决过程,提出了一种基于标签的说话人判决模型实现方案。该设计用FPGA实现,可对多路电话信道说话人进行实时判决识别。     

基于CHMM孤立词识别系统的快速实现

介绍了基于连续隐含马尔可夫模型(CHMM)的非特定人孤立词语音识别系统.根据分析该系统计算复杂度,提出了一系列的优化方法,通过MATLAB平台下的研究实验数据表明,优化后的语音识别系统与传统CHMM语音识别系统对比,计算时间是传统CHMM系统的9.97%,而识别率仅从传统CHMM系统的94%下降到91.3%.     

Design and implementation of a framework for building distributed smart object systems

A smart object system encompasses the synergy between computationally augmented everyday objects and external applications. This paper presents a software framework for building smart object systems following a declarative programming approach centered around custom written documents that glue the smart objects together. More specifically, in the proposed framework, applications’ requirements and smart objects’ services are objectified through structured documents. A runtime infrastructure provides the spontaneous federation between smart objects and applications through structural type matching of these documents. There are three primary advantages of our approach: firstly, it allows developers to write applications in a generic way without prior knowledge of the smart objects that could be used by the applications. Secondly, smart object management (locating, accessing, etc.) issues are completely handled by the infrastructure; thus application development becomes rapid and simple. Finally, the programming abstraction used in the framework allows extension of functionalities of smart objects and applications very easily. We describe an implemented prototype of our framework and show examples of its use in a real life scenario to illustrate its feasibility.     

基于FPGA动态重构的快速车牌识别系统

FPGA有限的片上资源限制了复杂神经网络的实现,通过动态部分重构技术,完成2种软硬件方案设计。与纯软件方式相比,网络执行速度提高了3倍以上;自制车牌字符数据集,在Caffe网络框架下实现了99. 45%的训练精度;最终,基于PYNQ—Z1开发板,完成了快速车牌识别系统的设计。     

基于视觉的移动机器人可通行区域识别研究综述

实现实时准确的可通行区域识别,是户外环境下移动机器人导航的重要组成部分。对基于视觉的移动机器人可通行区域识别研究进行了综述,首先介绍了移动机器人视觉导航常用的视觉系统,并从障碍物检测、地形分类两个方面介绍了该问题研究的进展,最后对该领域的技术发展趋势进行了探讨。     

视频监控系统中的图像分割算法综述*

视频监控系统在智能安防、人机交互、交通、娱乐、军事等领域有着广泛的应用空间,是近来研究热点之一。目标的分割是视频监控系统中的首要任务,其分割的有效性对于后续的目标识别、跟踪、行为理解等处理至关重要。从时间分割法和空间分割法的分类角度出发,对国内外的视频监控研究中的若干目标分割方法进行了归纳总结。     

基于立体视觉的非穿戴指势识别方法

基于彩色图像中红、绿、蓝3分量强度在阴影区域存在差异,根据小波变换在时域和空域均具有优异的局部化特征,结合背景差分,进行小波多尺度变换,提取视频指势对象,所提方法不需场景学习与训练、手工校正及先验假设等信息,可克服动态场景变化、阴影、噪声干扰等影响,具有强的鲁棒性.基于人类生物结构特征,采用不易遮挡和不受人脸朝向、姿态、光照变化等影响的头项特征代替人眼特征,保证了人机交互活动的自由性和自然性,且提高了人机交互的时效性.融合手指尖特征和手臂中心轴线及其外极线的多几何约束策略,采用求解反对应方法.确保手指特征匹配对应的正确性.通过实验验证,证实了上述方法有效、可行,可应用于实时、非穿戴的自然指势视觉3维人机交互中.     

Fast object recognition using dynamic programming from combination of salient line groups

This paper presents a new method of grouping and matching line segments to recognize objects. We propose a dynamic programming-based formulation extracting salient line patterns by defining a robust and stable geometric representation that is based on perceptual organizations. As the endpoint proximity, we detect several junctions from image lines. We then search for junction groups by using the collinear constraint between the junctions. Junction groups similar to the model are searched in the scene, based on a local comparison. A DP-based search algorithm reduces the time complexity for the search of the model lines in the scene. The system is able to find reasonable line groups in a short time.     

FPGA平台实现基于遗传算法的图像识别的研究

利用模板匹配方法,采用基于遗传算法的图像识别技术,完成了对图像目标识别的算法验证。在此基础上进行了基于该算法的图像识别系统的FPGA实现,并在相关验证平台进行了硬件仿真与时序分析。实验结果表明,所设计的图像识别电路具有较高的识别精度和较快的识别速度。     

基于小波分解的高温物体图像快速识别方法*

摘要：提出了一种基于小波分解和多分辨率分析的高温图像快速识别方法。首先通过小波分解降低原始图像直方图的灰度级,从而减小了分割阈值的搜索范围,再结合最大类间方差法计算分割阈值,这样可以快速找到分割阈值。实验结果表明,该方法能够显著提高分割速度并准确分割出目标图像。     

Aspect graphs for three‐dimensional object recognition machine vision systems

The purpose of this research is to seek evidence for viewer‐centered (especially aspect‐graph‐based) visual processing in the elementary task of object understanding. Two homologous, bilaterally symmetrical three‐dimensional (3‐D) objects have been employed that differ in that one is based on parts with flat surfaces and the other on parts with curved surfaces. The following procedure has been followed, separately for each object. In the training (saturated free inspection and manipulation) phase, a location (identical for both objects) of the object is marked with a red strip and the subjects' task is to memorize the object structure as well as the position of the strip. In the test phase, two‐dimensional views of the object without the strip are presented and the subjects' task is to determine whether the previously marked location should be visible or invisible in the particular view. Findings have been found consistent with an aspect‐graph‐based 3‐D object representation: (a) the reaction times and errors show characteristic dependencies on viewpoint; (b) a number of views (corresponding to certain aspects and aspect transitions of the aspect graph) consistently produce faster and more accurate recognition; (c) the differences in the aspect graphs of the two objects are reflected in differing patterns of reaction times and errors; furthermore; (d) the subjects impose a standard orientation on the objects, whereby a strong inversion effect is observed; and (e) performance varies in a similar way for both objects as a function of tilt. It is concluded that object understanding is viewpoint dependent, that is, based on a number of views. The characteristics of the views found to be most important for object understanding can be employed for creating efficient 3‐D object recognition machine vision systems. © 2005 Wiley Periodicals, Inc. Int J Int Syst 20: 47–72, 2005.     

基于FPGA和CNN的汉字识别系统的设计与实现

近年来,卷积神经网络在图像处理方面得到了广泛应用,然而其存在计算复杂,移动端资源有限,无法存储过多数据、进行大规模计算等缺点。提出一种基于CNN的汉字识别系统的硬件实现方法。在TensorFlow框架下用casia数据集训练出20个常用汉字的CNN网络架构,测试集识别率达98. 36%,并采用卷积核复用、定点化等方法降低资源消耗,在FPGA上搭建优化后的CNN。最后,将摄像头实时采集的图片输入到上述CNN,实现硬件端的汉字识别。实验结果表明,在结构简化、速度相较CPU提高6. 76倍的同时,在FPGA上所构建的CNN达到几乎无损的97. 58%的准确率。     

Evidential reasoning for object recognition

The authors present a framework to guide development of evidential reasoning in object recognition systems. Principles of evidential reasoning processes for open-world object recognition are proposed and applied to build evidential reasoning capabilities. The principles summarize research and findings by the authors up through the mid-1990s, including seminal results in object-centered computer vision, figure-ground discrimination, and the application of hierarchical Bayesian inference, Bayesian networks, and decision graphs to evidential reasoning for object recognition.