The Federal Railroad Administration (FRA)’s Web Based Accident Prediction System (WBAPS) is used by federal, state and local agencies to get a preliminary idea on safety at a rail-highway grade crossing. It is an int...The Federal Railroad Administration (FRA)’s Web Based Accident Prediction System (WBAPS) is used by federal, state and local agencies to get a preliminary idea on safety at a rail-highway grade crossing. It is an interactive and user-friendly tool used to make funding decisions. WBAPS is almost three decades old and involves a three-step approach making it difficult to interpret the contribution of the variables included in the model. It also does not directly account for regional/local developments and technological advancements pertaining to signals and signs implemented at rail-highway grade crossings. Further, characteristics of a rail-highway grade crossing vary by track class which is not explicitly considered by WBAPS. This research, therefore, examines and develops a method and models to estimate crashes at rail-highway grade crossings by track class using regional/local level data. The method and models developed for each track class as well as considering all track classes together are based on data for the state of North Carolina. Linear, as well as count models based on Poisson and Negative Binomial (NB) distributions, was tested for applicability. Negative binomial models were found to be the best fit for the data used in this research. Models for each track class have better goodness of fit statistics compared to the model considering data for all track classes together. This is primarily because traffic, design, and operational characteristics at rail-highway grade crossings are different for each track class. The findings from statistical models in this research are supported by model validation.展开更多
为准确实时跟踪羊只目标,进行疾病异常预警,实现奶山羊精细化养殖,本文基于DiMP跟踪模型,利用奶山羊跟踪对象单一且图像样本丰富的特点,结合迁移学习和类特定融合方法,设计了一种类特定的奶山羊目标跟踪模型,能够有效克服DiMP算法在跟...为准确实时跟踪羊只目标,进行疾病异常预警,实现奶山羊精细化养殖,本文基于DiMP跟踪模型,利用奶山羊跟踪对象单一且图像样本丰富的特点,结合迁移学习和类特定融合方法,设计了一种类特定的奶山羊目标跟踪模型,能够有效克服DiMP算法在跟踪类特定目标时定位精度不足的缺点。利用构建的奶山羊视频跟踪数据训练集对跟踪算法进行迁移训练,加快模型收敛速度,使评估网络预测出的边界框更贴合奶山羊真实框的位置和尺寸。在线跟踪阶段,针对目标模板仅采用第1帧特征制作整个序列的调制向量,导致该调制向量相对整个跟踪阶段特征不具代表性,与后续帧差异大的缺点,使用训练集制作包含奶山羊各种姿态的类调制向量,以指数消融方式更新奶山羊类调制向量与第1帧调制向量间的比重,增强边界框回归任务中的奶山羊特征与背景的判别性。提出的算法在测试集上的AUC(Area under curve)和精准度(Precision)分别为76.20%和60.19%,比DiMP方法分别提升6.17、14.18个百分点,跟踪速度为30 f/s,满足实时跟踪的要求。实验结果表明,提出的类特定奶山羊目标跟踪方法可用于监测复杂场景下奶山羊的运动,为奶山羊精细化管理提供了技术支持。展开更多
基于分类的跟踪算法成为当前目标跟踪的研究热点.首先把跟踪问题看成是一个目标和背景的二分类问题,根据每一帧的正负样本数据训练SVM分类器,通过分类器的分类概率值确定目标位置.然而,采集正负样本边界的那些样本很容易出现异常点,当...基于分类的跟踪算法成为当前目标跟踪的研究热点.首先把跟踪问题看成是一个目标和背景的二分类问题,根据每一帧的正负样本数据训练SVM分类器,通过分类器的分类概率值确定目标位置.然而,采集正负样本边界的那些样本很容易出现异常点,当把它们作为目标的下一帧位置时将会出现严重的跟踪漂移问题.本文在此基础上提出一种基于单类支持向量机(One-class support vector machine)的目标跟踪算法,基于One-class SVM分类能有效地排除其他类的干扰,有效地防止异常样本的出现.并结合加权多示例采样方法,使得每个采样样本会根据不同的权值对于分类器的贡献而不同.实验结果表明本文改进跟踪方法的鲁棒性.展开更多
文摘The Federal Railroad Administration (FRA)’s Web Based Accident Prediction System (WBAPS) is used by federal, state and local agencies to get a preliminary idea on safety at a rail-highway grade crossing. It is an interactive and user-friendly tool used to make funding decisions. WBAPS is almost three decades old and involves a three-step approach making it difficult to interpret the contribution of the variables included in the model. It also does not directly account for regional/local developments and technological advancements pertaining to signals and signs implemented at rail-highway grade crossings. Further, characteristics of a rail-highway grade crossing vary by track class which is not explicitly considered by WBAPS. This research, therefore, examines and develops a method and models to estimate crashes at rail-highway grade crossings by track class using regional/local level data. The method and models developed for each track class as well as considering all track classes together are based on data for the state of North Carolina. Linear, as well as count models based on Poisson and Negative Binomial (NB) distributions, was tested for applicability. Negative binomial models were found to be the best fit for the data used in this research. Models for each track class have better goodness of fit statistics compared to the model considering data for all track classes together. This is primarily because traffic, design, and operational characteristics at rail-highway grade crossings are different for each track class. The findings from statistical models in this research are supported by model validation.
文摘为准确实时跟踪羊只目标,进行疾病异常预警,实现奶山羊精细化养殖,本文基于DiMP跟踪模型,利用奶山羊跟踪对象单一且图像样本丰富的特点,结合迁移学习和类特定融合方法,设计了一种类特定的奶山羊目标跟踪模型,能够有效克服DiMP算法在跟踪类特定目标时定位精度不足的缺点。利用构建的奶山羊视频跟踪数据训练集对跟踪算法进行迁移训练,加快模型收敛速度,使评估网络预测出的边界框更贴合奶山羊真实框的位置和尺寸。在线跟踪阶段,针对目标模板仅采用第1帧特征制作整个序列的调制向量,导致该调制向量相对整个跟踪阶段特征不具代表性,与后续帧差异大的缺点,使用训练集制作包含奶山羊各种姿态的类调制向量,以指数消融方式更新奶山羊类调制向量与第1帧调制向量间的比重,增强边界框回归任务中的奶山羊特征与背景的判别性。提出的算法在测试集上的AUC(Area under curve)和精准度(Precision)分别为76.20%和60.19%,比DiMP方法分别提升6.17、14.18个百分点,跟踪速度为30 f/s,满足实时跟踪的要求。实验结果表明,提出的类特定奶山羊目标跟踪方法可用于监测复杂场景下奶山羊的运动,为奶山羊精细化管理提供了技术支持。
文摘基于分类的跟踪算法成为当前目标跟踪的研究热点.首先把跟踪问题看成是一个目标和背景的二分类问题,根据每一帧的正负样本数据训练SVM分类器,通过分类器的分类概率值确定目标位置.然而,采集正负样本边界的那些样本很容易出现异常点,当把它们作为目标的下一帧位置时将会出现严重的跟踪漂移问题.本文在此基础上提出一种基于单类支持向量机(One-class support vector machine)的目标跟踪算法,基于One-class SVM分类能有效地排除其他类的干扰,有效地防止异常样本的出现.并结合加权多示例采样方法,使得每个采样样本会根据不同的权值对于分类器的贡献而不同.实验结果表明本文改进跟踪方法的鲁棒性.
