基于DSP的前方车辆位置关系辨识研究

为提高前方车辆位置关系辨识效能,提出利用基于统计特性的图像识别算法辨识前方车辆位置关系。通过车载CCD实时获取道路图像信息,使用最小二乘法进行车道线拟合,结合道路图像同侧上、下车道标识线的斜率关系判定道路线形,以车道中线标定前方车辆位置;基于图像统计特性进行前方车辆识别,计算车辆标识点至车道中线的距离,通过与设定的阈值进行比较来确定前方车辆与自车的位置关系。实验结果表明,该算法能够有效降低由树阴和路面油污对辨识所造成的影响,抗干扰能力强,对不同曲率的道路具有良好的适应性。     

基于计算机视觉的前方车辆检测与测距系统设计

随着我国私家车数量的增多,公路行驶道路安全成为社会关注的主要问题.而前方车辆距离检测系统的搭建,能够完成前方车辆检测与智能汽车导航工作.探讨了计算机视觉的前方车辆检测与测距系统设计,通过分析基于前方车身底部阴影的检测算法,阐述了前方车辆检测与测距系统的设计与应用.     

基于改进的FAST R-CNN的前方车辆检测研究

目前,前方车辆检测的研究主要通过机器学习的方法,然而其难以解决遮挡和误检的问题。在这种背景下,使用深度学习的方法检测前方车辆更为有效。首先采用了选择性搜索方法获得样本图像的候选区域,然后使用改进的FAST R-CNN训练网络模型,检测道路前方车辆。已在KITTI车辆公共数据集上对该方法进行了测试,实验结果表明,所提方法的检测率高于CNN直接检测的结果,很大程度上解决了遮挡和误检的问题。而且,与先提取Harr-Like特征然后利用Adaptive Boosting分类器的算法相比,该方法在TSD-MAX交通场景数据库测试中实现了较高的性能。结果表明,该方法提高了车辆检测的准确性和鲁棒性。     

基于多尺度融合和注意力的前方车辆检测方法

论文提出了一种基于多尺度特征融合和通道注意力机制的神经网络方法,用于精确地检测前方车辆。该文提出的MSCA-Y模型是在YOLOv4的基础上结合了两个关键的见解：1）提出了一个高效的多尺度特征融合网络。在充分利用主干网络的特征表示能力的基础上,有效融合多尺度特征图的空间信息,以增强检测微型车辆的性能;2）引入了通道注意力机制。通过加强对车辆各种姿态上关键部位的特征关注度,以进一步在复杂检测任务中获得更好的检测性能。为了证明方法的有效性,论文在KITTI数据集上评估模型。实验表明,该方法在KITTI数据集的基准上取得了较高的mAP（在困难子集中达到了87.12%）。     

基于多地磁传感器的车辆位置检测系统设计

系统通过在路面下安装多个地磁传感器检测车辆在路面的实时位置。地磁传感器与微处理器集成为信号处理模块,信号处理模块采集地磁传感器数据并在滤波后发送给数据处理模块。数据处理模块通过车辆经过路面时地磁传感器数据的变化来分析计算得出车辆的位置。通过对实验测得数据进行曲线拟合得出车辆位置与传感器差值之间的关系,进而推导出车辆的位置,其两轴精度达到10 cm。此外,设计的系统还可以测得车辆的速度和前进方向,并有望在智能公路方面应用于自动驾驶。     

结合目标检测与双目视觉的三维车辆姿态检测

随着城市规模的发展,车辆的需求在与日俱增,同时对自动驾驶技术的需求也在不断提高.为了增强自动驾驶系统对路面车辆的信息掌握能力,提出一种车辆姿态检测方法.首先利用基于深度学习的目标检测方法获取车辆在二维图片上的信息,结合深度相机利用双目视觉获取车辆的关键三维空间信息;然后综合二维与三维信息建立三维空间坐标,经过计算后实现车辆的三维边框绘制,绘制的三维边框能辅助区分出车辆在空间上的方位.文中方法为端对端方法,不需要其他额外的输入信息,能够实时展示在相机中.实验结果表明,该方法针对常见的路面停车场景有较好的识别效果,对自动驾驶系统有较好的辅助作用;对比目前流行的三维边框计算方法也展示了其准确性.     

基于集成学习的入侵检测方法

入侵检测是近年来网络安全研究的热点。利用多分类器技术,研究了基于集成学习的入侵检测方法。应用Bootstrap技术生成分类器个体,为了提高分类器的差异性,应用聚类技术对分类器进行聚类,在相应的聚类结果中选取不同的分类器个体,并选择不同的融合方法对分类结果进行融合。针对入侵检测数据的实验表明了该集成技术的有效性。     

基于互信息加权集成迁移学习的入侵检测方法

入侵检测系统（IDS）已成为网络安全体系结构中的必要组成部分。在面对现代网络安全需求时,现有的入侵检测方法的可行性和持续性仍然存在提高空间,主要体现在更早地发现入侵威胁和提高入侵检测系统的检测精准度,为此提出一种基于互信息加权的集成迁移学习（ETL）入侵检测方法。首先,通过迁移策略对多组特征集进行建模;然后,使用互信息度量在迁移模型下特征集在不同域中的数据分布;最后,根据度量值对多个迁移模型进行集成加权,得到集成迁移模型。该方法通过学习新环境下的少量有标记样本和以往环境下的大量有标记样本的知识,可以建立效果优于传统非集成、非迁移的入侵检测模型。使用基准NSL-KDD数据集对该方法进行评估,实验结果表明,所提方法具有良好的收敛性能,并提高了入侵检测的精准率。     

基于视频的车辆发现模块的研究与实现

针对现有的智能交通系统的车辆发现模块的不足,提出了车辆发现的一整套解决方案。综合使用了已有的视频处理技术及方法,背景差分法、帧间差分法、检测线方法,以及自适应边框的车辆精确定位等方法,设计实现了智能交通系统的车辆发现模块。设计了三种不同的真实场景的实验。通过对结果的分析表明,系统的车辆发现模块对于车辆的识别度高达90%以上,在一般的设备下流畅运行,能够满足车辆发现应用对系统的实时性的要求,并且适应性也较好。     

智能车中基于单目视觉的前车检测和跟踪

提出了一个改进的单目视觉方法,用于智能车在结构化公路环境下准确检测和跟踪前方车辆。该方法先利用图像灰度梯度检测前车,剔除可能的虚检测,建立新目标的二维模型;然后用卡尔曼滤波方法预测下一帧的目标位置,在预测位置附近用边缘投影方法定位目标;设计了一种新的四因素似然度函数,验证跟踪结果与检测结果的匹配度,当跟踪失败时,重新检测前车。利用长图像序列PETS2001进行实验,结果表明该方法可以有效的检测和跟踪本车车道前方视野中的车辆障碍物,为智能车的防撞预警和控制系统提供可靠信息。     

基于对抗学习和全局知识信息的关系检测技术研究

针对现有的知识库关系检测任务对于一些不可见关系无法做到准确的向量表示而出现词汇溢出的问题,提出了基于对抗学习和全局知识信息的关系检测模型。该模型使用对抗学习对知识库关系表示模型进行特征强化,使用TransH(translating on hyperplanes)模型提取全局知识信息,同时通过联合训练,将全局知识信息融合进关系表示模型中,进一步提升关系模型的表示能力。实验结果表明,提出的融合模型对于关系检测效果有一定的提升,并且缓解了词汇溢出的问题。     

一种基于改进码本的车辆检测与跟踪方法

为了解决固定摄像机下车辆跟踪过程中阴影对检测的影响,提出一种改进型码本模型的车辆检测方法。该方法直接对YUV空间的车辆序列进行处理,将采样到的背景值聚类成码本,对于新输入的像素值与其对应位置的码本作比较判断,提取出前景区域。车辆跟踪中采用Kalman预测的方法来处理车辆遮挡问题。实验结果表明,本算法可以从复杂交通场景图像序列中快速有效地检测出运动目标,能较好地处理阴影、高亮、遮挡和背景变化等问题,且计算复杂度小,能满足实时跟踪的需要。     

基于级联宽度学习的疲劳驾驶检测

为减少因疲劳驾驶引发的交通事故,提出融合多参数的驾驶员疲劳检测算法。用渐进校准网络(PCN)检测人脸图像,通过基于CNN的回归模型定位人脸关键点;根据关键点坐标和面部器官的分布规律提取眼睛和嘴部图像,用宽度学习系统(BLS)分别识别眼睛与嘴巴的状态;将眼睛、嘴巴和头部状态的时序序列送入二级宽度网络对司机的状态进行判别。实验结果表明,该算法的疲劳检测准确率为94.9%,单帧检测时间52.43ms。     

基于度量学习的行人检测算法

针对拥挤场景下行人漏检率较高的问题,设计了新的类平衡策略.其次,采用度量学习方法改进目前的行人语义提取效果,并设计了新的距离度量方法.最后,结合提取的行人语义信息设计了新的非极大值抑制算法.在行人检测数据集CityPersons和CrowdHuman上,与目前的行人检测器进行对比,效果优于目前最优无锚框的行人检测器,同时也证明了度量学习方法在行人检测中的有效性.     

Abnormal visual event detection based on multi-instance learning and autoregressive integrated moving average model in edge-based Smart City surveillance

The abnormal visual event detection is an important subject in Smart City surveillance where a lot of data can be processed locally in edge computing environment. Real-time and detection effectiveness are critical in such an edge environment. In this paper, we propose an abnormal event detection approach based on multi-instance learning and autoregressive integrated moving average model for video surveillance of crowded scenes in urban public places, focusing on real-time and detection effectiveness. We propose an unsupervised method for abnormal event detection by combining multi-instance visual feature selection and the autoregressive integrated moving average model. In the proposed method, each video clip is modeled as a visual feature bag containing several subvideo clips, each of which is regarded as an instance. The time-transform characteristics of the optical flow characteristics within each subvideo clip are considered as a visual feature instance, and time-series modeling is carried out for multiple visual feature instances related to all subvideo clips in a surveillance video clip. The abnormal events in each surveillance video clip are detected using the multi-instance fusion method. This approach is verified on publically available urban surveillance video datasets and compared with state-of-the-art alternatives. Experimental results demonstrate that the proposed method has better abnormal event detection performance for crowded scene of urban public places with an edge environment.     

Partial mutual information based input variable selection for supervised learning approaches to voice activity detection

The paper presents a novel approach for voice activity detection. The main idea behind the presented approach is to use, next to the likelihood ratio of a statistical model-based voice activity detector, a set of informative distinct features in order to, via a supervised learning approach, enhance the detection performance. The statistical model-based voice activity detector, which is chosen based on the comparison to other similar detectors in an earlier work, models the spectral envelope of the signal and we derive the likelihood ratio thereof. Furthermore, the likelihood ratio together with 70 other various features was meticulously analyzed with an input variable selection algorithm based on partial mutual information. The resulting analysis produced a 13 element reduced input vector which when compared to the full input vector did not undermine the detector performance. The evaluation is performed on a speech corpus consisting of recordings made by six different speakers, which were corrupted with three different types of noises and noise levels. In the end, we tested three different supervised learning algorithms for the task, namely, support vector machine, Boost, and artificial neural networks. The experimental analysis was performed by 10-fold cross-validation due to which threshold averaged receiver operating characteristics curves were constructed. Also, the area under the curve score and Matthew's correlation coefficient were calculated for both the three supervised learning classifiers and the statistical model-based voice activity detector. The results showed that the classifier with the reduced input vector significantly outperformed the standalone detector based on the likelihood ratio, and that among the three classifiers, Boost showed the most consistent performance.     

Community structure detection in social networks based on dictionary learning

Discovering community structures is a fundamental problem concerning how to understand the topology and the functions of complex network. In this paper, we propose how to apply dictionary learning algorithm to community structure detection. We present a new dictionary learning algorithm and systematically compare it with other state-of-the-art models/algorithms. The results show that the proposed algorithm is highly effectively at finding the community structures in both synthetic datasets, including three types of data structures, and real world networks coming from different areas.     

Transfer learning and information retrieval applied to fall detection

Detecting falls in the elderly population is a very important issue that is related with the time of recovery. This study focuses on using wearable smart watches to monitor the movements of the user in order to detect patterns that might be related to fall events. The proposed solution explores Symbolic Aggregate approXimation (SAX) Time Series representation, together with two information retrieval techniques enriched with transfer learning (TL). The solution is user centred; that is, a model is developed for each specific user. Basically, the fall detection approach makes use of a finite-state machine to detect peaks; the time series window embedding these peaks are represented using SAX. Assuming the data from the public fall detection data sets are valid, a dictionary is prepared using the most relevant words. This dictionary is then introduced as previous knowledge to an online learning classifier that is trained with normal activities of daily living. The two classifiers are evaluated and compared with two classical approaches. Before this comparison, two clustering approaches are studied to produce the bag of relevant words. A complete experimentation is included, which makes use of several publicly available data sets and also with a data set developed by the research group. Comparisons are performed for all the data sets, showing how the TL stage empowers the classifier. The results show that this solution produces high detection rates and at the same time performed similarly for all the individuals tested. Furthermore, the positive effects of TL in this context are clearly remarked.