在交通场景中采用一些预警措施能够有效地减少交通事故发生。例如,对车辆轨迹进行跟踪并预测车辆的驾驶行为,就是一个常用的预警方法。在对车辆进行跟踪的过程中,数据关联是很重要的部分,它可以对车辆的观测点和轨迹进行关联,从而更新...在交通场景中采用一些预警措施能够有效地减少交通事故发生。例如,对车辆轨迹进行跟踪并预测车辆的驾驶行为,就是一个常用的预警方法。在对车辆进行跟踪的过程中,数据关联是很重要的部分,它可以对车辆的观测点和轨迹进行关联,从而更新车辆的轨迹,完成跟踪过程。在此背景下,提出了一种新的数据关联算法,即k近邻联合概率数据关联算法(k Nearest Neighbor-Joint Probability Data Association,kNN-JPDA)。实验结果表明,该算法能够较好地解决在交通场景下车辆数据的数据关联问题,在精度以及运行效率方面都有所提高。展开更多
目的车辆多目标跟踪是智能交通领域关键技术,其性能对车辆轨迹分析和异常行为鉴别有显著影响。然而,车辆多目标跟踪常受外部光照、道路环境因素影响,车辆远近尺度变化以及相互遮挡等干扰,导致远处车辆漏检或车辆身份切换(ID switch,IDs...目的车辆多目标跟踪是智能交通领域关键技术,其性能对车辆轨迹分析和异常行为鉴别有显著影响。然而,车辆多目标跟踪常受外部光照、道路环境因素影响,车辆远近尺度变化以及相互遮挡等干扰,导致远处车辆漏检或车辆身份切换(ID switch,IDs)问题。本文提出短时记忆与CenterTrack的车辆多目标跟踪,提升车辆多目标跟踪准确度(multiple object tracking accuracy,MOTA),改善算法的适应性。方法利用小样本扩增增加远处小目标车辆训练样本数;通过增加的样本重新训练CenterTrack确定车辆位置及车辆在相邻帧之间的中心位移量;当待关联轨迹与检测目标匹配失败时通过轨迹运动信息预测将来的位置;利用短时记忆将待关联轨迹按丢失时间长短分级与待匹配检测关联以减少跟踪车辆IDs。结果在交通监控车辆多目标跟踪数据集UA-DETRAC(University at Albany detection and tracking)构建的5个测试序列数据中,本文方法在维持CenterTrack优势的同时,对其表现不佳的场景获得近30%的提升,与YOLOv4-DeepSort(you only look once—simple online and realtime tracking with deep association metric)相比,4种场景均获得近10%的提升,效果显著。Sherbrooke数据集的测试结果,本文方法同样获得了性能提升。结论本文扩增了远处小目标车辆训练样本,缓解了远处小目标与近处大目标存在的样本不均衡,提高了算法对远处小目标车辆的检测能力,同时短时记忆维持关联失败的轨迹运动信息并分级匹配检测目标,降低了算法对跟踪车辆的IDs,综合提高了MOTA。展开更多
Women’s health is important for society.Despite the known biological and sex-related factors influencing the risk of diseases among women,the network of the full spectrum of diseases in women is underexplored.This st...Women’s health is important for society.Despite the known biological and sex-related factors influencing the risk of diseases among women,the network of the full spectrum of diseases in women is underexplored.This study aimed to systematically examine the women-specific temporal pattern(trajectory)of the disease network,including the role of baseline physical examination indexes,and blood and urine biomarkers.In the UK Biobank study,502,650 participants entered the cohort from 2006 to 2010,and were followed up until 2019 to identify disease incidence via linkage to the patient registers.For those diseases with increased risk among women,conditional logistic regression models were used to estimate odds ratios(ORs),and the binomial test of direction was further used to build disease trajectories.Among 301 diseases,82 diseases in women had ORs>1.2 and p<0.00017 when compared to men,involving mainly diseases in the endocrine,skeletal and digestive systems.Diseases with the highest ORs included breast diseases,osteoporosis,hyperthyroidism,and deformity of the toes.The biomarker and disease trajectories suggested estradiol as a risk predictor for breast cancer,while a high percentage of reticulocyte,body mass index and waist circumference were associated with an increased risk of upper-limb neuropathy.In addition,the risk of cholelithiasis was increased in women diagnosed with dyspepsia and diaphragmatic hernia.In conclusion,women are at an increased risk of endocrine,skeletal and digestive diseases.The biomarker and disease trajectories in women suggested key pathways to a range of adverse outcomes downstream,which may shed light on promising targets for early detection and prevention of these diseases.展开更多
文摘在交通场景中采用一些预警措施能够有效地减少交通事故发生。例如,对车辆轨迹进行跟踪并预测车辆的驾驶行为,就是一个常用的预警方法。在对车辆进行跟踪的过程中,数据关联是很重要的部分,它可以对车辆的观测点和轨迹进行关联,从而更新车辆的轨迹,完成跟踪过程。在此背景下,提出了一种新的数据关联算法,即k近邻联合概率数据关联算法(k Nearest Neighbor-Joint Probability Data Association,kNN-JPDA)。实验结果表明,该算法能够较好地解决在交通场景下车辆数据的数据关联问题,在精度以及运行效率方面都有所提高。
文摘目的车辆多目标跟踪是智能交通领域关键技术,其性能对车辆轨迹分析和异常行为鉴别有显著影响。然而,车辆多目标跟踪常受外部光照、道路环境因素影响,车辆远近尺度变化以及相互遮挡等干扰,导致远处车辆漏检或车辆身份切换(ID switch,IDs)问题。本文提出短时记忆与CenterTrack的车辆多目标跟踪,提升车辆多目标跟踪准确度(multiple object tracking accuracy,MOTA),改善算法的适应性。方法利用小样本扩增增加远处小目标车辆训练样本数;通过增加的样本重新训练CenterTrack确定车辆位置及车辆在相邻帧之间的中心位移量;当待关联轨迹与检测目标匹配失败时通过轨迹运动信息预测将来的位置;利用短时记忆将待关联轨迹按丢失时间长短分级与待匹配检测关联以减少跟踪车辆IDs。结果在交通监控车辆多目标跟踪数据集UA-DETRAC(University at Albany detection and tracking)构建的5个测试序列数据中,本文方法在维持CenterTrack优势的同时,对其表现不佳的场景获得近30%的提升,与YOLOv4-DeepSort(you only look once—simple online and realtime tracking with deep association metric)相比,4种场景均获得近10%的提升,效果显著。Sherbrooke数据集的测试结果,本文方法同样获得了性能提升。结论本文扩增了远处小目标车辆训练样本,缓解了远处小目标与近处大目标存在的样本不均衡,提高了算法对远处小目标车辆的检测能力,同时短时记忆维持关联失败的轨迹运动信息并分级匹配检测目标,降低了算法对跟踪车辆的IDs,综合提高了MOTA。
基金Open access funding is provided by Karolinska Institute.This work is supported by the Grant of Science and Technology of Fujian,China[2019L3006,2020L3009]HY is supported by the Natural Science Foundation of Fujian Province[grant no:2021J01721]+2 种基金the Startup Fund for High-level Talents of Fujian Medical University[grant no:XRCZX2020007]Startup Fund for Scientific Research,Fujian Medical University[grant no:2019QH1002]Laboratory Construction Program of Fujian Medical University[grant no:1100160208].
文摘Women’s health is important for society.Despite the known biological and sex-related factors influencing the risk of diseases among women,the network of the full spectrum of diseases in women is underexplored.This study aimed to systematically examine the women-specific temporal pattern(trajectory)of the disease network,including the role of baseline physical examination indexes,and blood and urine biomarkers.In the UK Biobank study,502,650 participants entered the cohort from 2006 to 2010,and were followed up until 2019 to identify disease incidence via linkage to the patient registers.For those diseases with increased risk among women,conditional logistic regression models were used to estimate odds ratios(ORs),and the binomial test of direction was further used to build disease trajectories.Among 301 diseases,82 diseases in women had ORs>1.2 and p<0.00017 when compared to men,involving mainly diseases in the endocrine,skeletal and digestive systems.Diseases with the highest ORs included breast diseases,osteoporosis,hyperthyroidism,and deformity of the toes.The biomarker and disease trajectories suggested estradiol as a risk predictor for breast cancer,while a high percentage of reticulocyte,body mass index and waist circumference were associated with an increased risk of upper-limb neuropathy.In addition,the risk of cholelithiasis was increased in women diagnosed with dyspepsia and diaphragmatic hernia.In conclusion,women are at an increased risk of endocrine,skeletal and digestive diseases.The biomarker and disease trajectories in women suggested key pathways to a range of adverse outcomes downstream,which may shed light on promising targets for early detection and prevention of these diseases.
