Predicting Object Dynamics From Visual Images Through Active Sensing Experiences

Prediction of dynamic features is an important task for determining the manipulation strategies of an object. This paper presents a technique for predicting dynamics of objects relative to the robot's motion from visual images. During the training phase, the authors use the recurrent neural network with parametric bias (RNNPB) to self-organize the dynamics of objects manipulated by the robot into the PB space. The acquired PB values, static images of objects and robot motor values are input into a hierarchical neural network to link the images to dynamic features (PB values). The neural network extracts prominent features that each induce object dynamics. For prediction of the motion sequence of an unknown object, the static image of the object and robot motor value are input into the neural network to calculate the PB values. By inputting the PB values into the closed loop RNNPB, the predicted movements of the object relative to the robot motion are calculated recursively. Experiments were conducted with the humanoid robot Robovie-IIs pushing objects at different heights. The results of the experiment predicting the dynamics of target objects proved that the technique is efficient for predicting the dynamics of the objects.     

Design of robotic visual servo control based on neural network and genetic algorithm

A new visual servo control scheme for a robotic manipulator is presented in this paper, where a back propagation (BP) neural network is used to make a direct transition from image feature to joint angles without requiring robot kinematics and camera calibration. To speed up the convergence and avoid local minimum of the neural network, this paper uses a genetic algorithm to find the optimal initial weights and thresholds and then uses the BP algorithm to train the neural network according to the data given. The proposed method can effectively combine the good global searching ability of genetic algorithms with the accurate local searching feature of BP neural network. The Simulink model for PUMA560 robot visual servo system based on the improved BP neural network is built with the Robotics Toolbox of Matlab. The simulation results indicate that the proposed method can accelerate convergence of the image errors and provide a simple and effective way of robot control.     

Combining multi-visual features for efficient indexing in a large image database

The optimized distance-based access methods currently available for multidimensional indexing in multimedia databases have been developed based on two major assumptions: a suitable distance function is known a priori and the dimensionality of the image features is low. It is not trivial to define a distance function that best mimics human visual perception regarding image similarity measurements. Reducing high-dimensional features in images using the popular principle component analysis (PCA) might not always be possible due to the non-linear correlations that may be present in the feature vectors. We propose in this paper a fast and robust hybrid method for non-linear dimensions reduction of composite image features for indexing in large image database. This method incorporates both the PCA and non-linear neural network techniques to reduce the dimensions of feature vectors so that an optimized access method can be applied. To incorporate human visual perception into our system, we also conducted experiments that involved a number of subjects classifying images into different classes for neural network training. We demonstrate that not only can our neural network system reduce the dimensions of the feature vectors, but that the reduced dimensional feature vectors can also be mapped to an optimized access method for fast and accurate indexing. Received 11 June 1998 / Accepted 25 July 2000 Published online: 13 February 2001     

基于场景理解与改进型BUG算法的移动机器人避障

针对现有移动机器人在视觉避障上存在的局限,将深度学习算法和路径规划技术相结合,提出了一种基于深层卷积神经网络和改进Bug算法的机器人避障方法;该方法采用多任务深度卷积神经网络提取道路图像特征,实现图像分类和语义分割任务;其次,基于语义分割结果构建栅格地图,并将图像分类结果与改进的Bug算法相结合,搜索出最优避障路径;同时,为降低冗余计算,设计了特征对比结构来对避免对重复计算的特征信息,保障机器人在实际应用中实时性;通过实验结果表明,所提方法有效的平衡了多视觉任务的精度与效率,并能准确规划出安全的避障路径,辅助机器人完成导航避障。     

基于图像矩与神经网络的机器人四自由度视觉伺服

针对传统的视觉伺服方法中图像几何特征的标记、提取与匹配过程复杂且通用性差等问题,本文提出了一种基于图像矩的机器人四自由度(4DOF)视觉伺服方法.首先建立了眼在手系统中图像矩与机器人位姿之间的非线性增量变换关系,为利用图像矩进行机器人视觉伺服控制提供了理论基础,然后在未对摄像机与手眼关系进行标定的情况下,利用反向传播(BP)神经网络的非线性映射特性设计了基于图像矩的机器人视觉伺服控制方案,最后用训练好的神经刚络进行了视觉伺服跟踪控制.实验结果表明基于本文算法可实现0.5 mm的位置与0.5°的姿态跟踪精度,验证了算法的的有效性与较好的伺服性能.     

基于改进RCE和RBF神经网络的静态手势识别

针对手势识别的手区域分割、手势特征提取和手势分类的三个过程,提出了一种新的静态手势识别方法。改进了传统的RCE神经网络用于手区域的分割,具有更高的运行速度和更强的抗噪能力。依Freeman链码方向提取手的边缘到掌心的距离作为手势的特征向量。将上一步得到的手势特征向量作为RBF神经网络的输入,进行网络的训练和分类。实验验证了该方法的有效性和可行性,并用其实现了人和仿人机器人的剪刀石头布的猜拳游戏。     

基于图像视觉伺服的二维运动优化控制

基于图像的视觉伺服方法,图像的变化直接解释为摄像机的运动,而不是直接对机械手末端实现笛卡尔速度控制,导致机械手的运动轨迹迂回,产生摄像机回退现象.针对这一问题,提出了将旋转和平移分离并先实现旋转的视觉伺服方案.该方案计算量小,系统响应时间短,解决了图像旋转和平移间的干扰,克服了传统基于图像视觉伺服产生的摄像机回退现象,实现了时间和路径的最优控制.并用传统IBVS的控制律和摄像机成像模型解释了回退现象的产生原因.二维运动仿真说明了提出方案的有效性.     

结合自底向上注意力机制和记忆网络的视觉问答模型

目的 现有大多数视觉问答模型均采用自上而下的视觉注意力机制,对图像内容无加权统一处理,无法更好地表征图像信息,且因为缺乏长期记忆模块,无法对信息进行长时间记忆存储,在推理答案过程中会造成有效信息丢失,从而预测出错误答案。为此,提出一种结合自底向上注意力机制和记忆网络的视觉问答模型,通过增强对图像内容的表示和记忆,提高视觉问答的准确率。方法 预训练一个目标检测模型提取图像中的目标和显著性区域作为图像特征,联合问题表示输入到记忆网络,记忆网络根据问题检索输入图像特征中的有用信息,并结合输入图像信息和问题表示进行多次迭代、更新,以生成最终的信息表示,最后融合记忆网络记忆的最终信息和问题表示,推测出正确答案。结果 在公开的大规模数据集VQA （visual question answering）v2.0上与现有主流算法进行比较实验和消融实验,结果表明,提出的模型在视觉问答任务中的准确率有显著提升,总体准确率为64.0%。与MCB（multimodal compact bilinear）算法相比,总体准确率提升了1.7%;与性能较好的VQA machine算法相比,总体准确率提升了1%,其中回答是/否、计数和其他类型问题的准确率分别提升了1.1%、3.4%和0.6%。整体性能优于其他对比算法,验证了提出算法的有效性。结论 本文提出的结合自底向上注意力机制和记忆网络的视觉问答模型,更符合人类的视觉注意力机制,并且在推理答案的过程中减少了信息丢失,有效提升了视觉问答的准确率。     

Self-organization of Dynamic Object Features Based on Bidirectional Training

This paper presents a method to self-organize object features that describe object dynamics using bidirectional training. The model is composed of a dynamics learning module and a feature extraction module. Recurrent Neural Network with Parametric Bias (RNNPB) is utilized for the dynamics learning module, learning and self-organizing the sequences of robot and object motions. A hierarchical neural network is linked to the input of RNNPB as the feature extraction module for self-organizing object features that describe the object motions. The two modules are simultaneously trained through bidirectional training using image and motion sequences acquired from the robot's active sensing with objects. Experiments are performed with the robot's pushing motion with a variety of objects to generate sliding, falling over, bouncing and rolling motions. The results have shown that the model is capable of self-organizing object dynamics based on the self-organized features.     

Supervised learning technique for a mobile robot controller in a visual line tracking task

This article deals with the development of learning methods for an intelligent control system for an autonomous mobile robot. On the basis of visual servoing, an approach to learning the skill of tracking colored guidelines is proposed. This approach utilizes a robust and adaptive image processing method to acquire features of the colored guidelines and convert them into the controller input. The supervised learning procedure and the neural network controller are discussed. The method of obtaining the learning data and training the neural network are described. Experimental results are presented at the end of the article. This work was presented, in part, at the Sixth International Symposium on Artificial Life and Robotics, Tokyo, Japan, January 15–17, 2001     

Neural network Reinforcement Learning for visual control of robot manipulators

It is known that most of the key problems in visual servo control of robots are related to the performance analysis of the system considering measurement and modeling errors. In this paper, the development and performance evaluation of a novel intelligent visual servo controller for a robot manipulator using neural network Reinforcement Learning is presented. By implementing machine learning techniques into the vision based control scheme, the robot is enabled to improve its performance online and to adapt to the changing conditions in the environment. Two different temporal difference algorithms (Q-learning and SARSA) coupled with neural networks are developed and tested through different visual control scenarios. A database of representative learning samples is employed so as to speed up the convergence of the neural network and real-time learning of robot behavior. Moreover, the visual servoing task is divided into two steps in order to ensure the visibility of the features: in the first step centering behavior of the robot is conducted using neural network Reinforcement Learning controller, while the second step involves switching control between the traditional Image Based Visual Servoing and the neural network Reinforcement Learning for enabling approaching behavior of the manipulator. The correction in robot motion is achieved with the definition of the areas of interest for the image features independently in both control steps. Various simulations are developed in order to present the robustness of the developed system regarding calibration error, modeling error, and image noise. In addition, a comparison with the traditional Image Based Visual Servoing is presented. Real world experiments on a robot manipulator with the low cost vision system demonstrate the effectiveness of the proposed approach.     

Visual quality recognition of nonwovens using generalized Gaussian density model and robust Bayesian neural network

This work is dedicated to develop an algorithm for the visual quality recognition of nonwoven materials, in which image analysis and neural network are involved in feature extraction and pattern recognition stage, respectively. During the feature extraction stage, each image is decomposed into four levels using the 9-7 bi-orthogonal wavelet base. Then the wavelet coefficients in each subband are independently modeled by the generalized Gaussian density (GGD) model to calculate the scale and shape parameters with maximum likelihood (ML) estimator as texture features. While for the recognition stage, the robust Bayesian neural network is employed to classify the 625 nonwoven samples into five visual quality grades, i.e., 125 samples for each grade. Finally, we carry out the outlier detection of the training set using the outlier probability and select the most suitable model structure and parameters from 40 Bayesian neural networks using the Occam's razor. When 18 relevant textural features are extracted for each sample based on the GGD model, the average recognition accuracy of the test set arranges from 88% to 98.4% according to the different number of the hidden neurons in the Bayesian neural network.     

仿人机器人视觉导航中的实时性运动模糊探测器设计

针对仿人机器人视觉导航系统的鲁棒性受到运动模糊制约的问题,提出一种基于运动模糊特征的实时性异常探测方法. 首先定量地分析运动模糊对视觉导航系统的负面影响,然后研究仿人机器人上图像的运动模糊规律,在此基础上对图像的运动模糊特征进行无参考的度量,随后采用无监督的异常探测技术,在探测框架下对时间序列上发生的图像运动模糊特征进行聚类分析,实时地召回数据流中的模糊异常,以增强机器人视觉导航系统对运动模糊的鲁棒性. 仿真实验和仿人机器人实验表明:针对国际公开的标准数据集和仿人机器人NAO数据集,方法具有良好的实时性（一次探测时间0.1s）和有效性（召回率98.5%,精确率90.7%）. 方法的探测框架对地面移动机器人亦具有较好的普适性和集成性,可方便地与视觉导航系统协同工作.     

多级卷积特征金字塔的细粒度食物图片识别

目的 食物图片具有结构多变、背景干扰大、类间差异小、类内差异大等特点,比普通细粒度图片的识别难度更大。目前在食物图片识别领域,食物图片的识别与分类仍存在精度低、泛化性差等问题。为了提高食物图片的识别与分类精度,充分利用食物图片的全局与局部细节信息,本文提出了一个多级卷积特征金字塔的细粒度食物图片识别模型。方法 本文模型从整体到局部逐级提取特征,将干扰较大的背景信息丢弃,仅针对食物目标区域提取特征。模型主要由食物特征提取网络、注意力区域定位网络和特征融合网格3部分组成,并采用3级食物特征提取网络的级联结构来实现特征由全局到局部的转移。此外,针对食物图片尺度变化大的特点,本文模型在每级食物特征提取网络中加入了特征金字塔结构,提高了模型对目标大小的鲁棒性。结果 本文模型在目前主流公开的食物图片数据集Food-101、ChineseFoodNet和Food-172上进行实验,分别获得了91.4%、82.8%、90.3%的Top-1正确率,与现有方法相比提高了1%~8%。结论 本文提出了一种多级卷积神经网络食物图片识别模型,可以自动定位食物图片区分度较大的区域,融合食物图片的全局与局部特征,实现了食物图片的细粒度识别,有效提高了食物图片的识别精度。实验结果表明,该模型在目前主流食物图片数据集上取得了最好的结果。     

TEXTNN—A MATLAB program for textural classification using neural networks

A new MATLAB code that provides tools to perform classification of textural images for applications in the geosciences is presented in this paper. The program, here coined as textural neural network (TEXTNN), comprises the computation of variogram maps in the frequency domain for specific lag distances in the neighborhood of a pixel. The result is then converted back to spatial domain, where directional or omni-directional semivariograms are extracted. Feature vectors are built with textural information composed of semivariance values at these lag distances and, moreover, with histogram measures of mean, standard deviation and weighted-rank fill ratio. This procedure is applied to a selected group of pixels or to all pixels in an image using a moving window. A feed-forward back-propagation neural network can then be designed and trained on feature vectors of predefined classes (training set). The training phase minimizes the mean-squared error on the training set. Additionally, at each iteration, the mean-squared error for every validation is assessed and a test set is evaluated. The program also calculates contingency matrices, global accuracy and kappa coefficient for the training, validation and test sets, allowing a quantitative appraisal of the predictive power of the neural network models. The interpreter is able to select the best model obtained from a k-fold cross-validation or to use a unique split-sample dataset for classification of all pixels in a given textural image. The performance of the algorithms and the end-user program were tested using synthetic images, orbital synthetic aperture radar (SAR) (RADARSAT) imagery for oil-seepage detection, and airborne, multi-polarized SAR imagery for geologic mapping, and the overall results are considered quite positive.     

Self-supervised learning for docking and target reaching

A robotic system using simple visual processing and controlled by neural networks is described. The robot performs docking and target reaching without prior geometric calibration of its components. All effects of control signals on the robot are learned by the controller through visual observation during a training period, and refined during actual operation. Minor changes in the system's configuration result in a brief period of degraded performance while the controller adapts to the new mappings.

It is shown that a neural network-based controller can perform rapidly and accurately, taking into account the non-linearities of various mapping functions. Such a controller is easy to train, tolerant of imprecise equipment configurations, and insensitive to camera perturbations following training. This method features real-time adaptivity to changes in mappings, and is simpler than traditional control techniques, which require the solution of the inverse perspective projection and inverse kinematics of the system.

Various operations including approaching, centering, paralleling, reaching and adjusting are performed by the robot as it navigates towards the target. The robot attempts to grasp targets that are sufficiently close, or approach them while avoiding collisions with obstacles.     

Towards selective attention: generating image features by learning a visuo-motor map

Robots require a form of visual attention to perform a wide range of tasks effectively. Existing approaches specify in advance the image features and attention control scheme required for a given robot to perform a specific task. However, to cope with different tasks in a dynamic environment, a robot should be able to construct its own attentional mechanisms. This paper presents a method that a robot can use to generating image features by learning a visuo-motor map. The robot constructs the visuo-motor map from training data, and the map constrains both the generation of image features and the estimation of state vectors. The resulting image features and state vectors are highly task-oriented. The learned mechanism is attentional in the sense that it determines what information to select from the image to perform a task. We examine robot experiments using the proposed method for indoor navigation and scoring soccer goals.     

基于BP神经网络的水果识别系统

近些年,计算机视觉发展迅速,在水果识别方向进行了广泛的应用和研究。本文设计基于BP神经网络的水果识别系统,选取生活中常见的三种水果:苹果、橘子、香蕉作为对象。首先,通过网络资源等搜集水果图像建立样本库;然后通过MATLAB对图像进行预处理,为后续的特征提取做好准备。水果特征的提取选择纹理、形状、颜色三种特征进行提取;同时在每种特征中选用不同的特征值作为特征向量。通过提取三种特征后输入到BP神经网络中进行训练、识别。经测试,识别的成功率可以达到93.18%,证明了可行性以及未来的可实用性。