基于深度学习的安全帽佩戴检测与跟踪

为了解决传统施工现场安全管理的弊端,减少因施工人员未佩戴安全帽造成的人员伤亡,本文提出一种基于深度学习的安全帽佩戴检测与跟踪方法。首先通过深度学习YOLOv3目标检测网络实现安全帽佩戴检测,进一步运用卡尔曼滤波器和KM算法实现多目标跟踪与计数。复杂施工现场的测试结果表明：网络模型的检测速度可达45 fps,平均精确度为93%,且未佩戴安全帽的查准率和查全率分别为97%和95%,基本能够实现安全帽佩戴情况的实时检测。     

改进YOLOv3的安全帽佩戴检测方法

针对在智能监控中安全帽佩戴检测准确率低和检测速率慢的问题,提出一种基于改进YOLOv3（You Only Look Once）的安全帽佩戴检测算法YOLOv3-WH。在YOLOv3算法的基础上改进网络结构,增大输入图像的尺度,使用深度可分离卷积结构替换Darknet-53传统卷积,减少特征的丢失,缩减模型参数,提升检测速率;使用多尺度特征检测,增加浅层检测尺度,添加4倍上采样特征融合结构,提高安全帽佩戴检测准确率;优化[K]-Means聚类算法,获取安全帽佩戴检测的先验框（anchor box）,按照预测尺度大小分配适合的anchor,提升模型训练和检测速率。实验结果表明YOLOv3-WH相比YOLOv3,每秒检测帧数（FPS）提高了64%,检测平均精确度（mAP）提高了6.5%,该算法在提升了安全帽佩戴检测速率的同时提升了检测的准确率,对安全帽佩戴检测具有一定的实用性。     

The effect of a helmet on cognitive performance is,at worst,marginal: A controlled laboratory study

The present study looked at the effect of a helmet on cognitive performance under demanding conditions, so that small effects would become more detectible. Nineteen participants underwent 30 min of continuous visual vigilance, tracking, and auditory vigilance (VTT + AVT), while seated in a warm environment (27.2 (±0.6) °C, humidity 41 (±1)%, and 0.5 (±0.1) m s⁻¹ wind speed). The participants wore a helmet in one session and no helmet in the other, in random order. Comfort and temperature perception were measured at the end of each session. Helmet-wearing was associated with reduced comfort (p = 0.001) and increased temperature perception (p < 0.001), compared to not wearing a helmet. Just one out of nine cognitive parameters showed a significant effect of helmet-wearing (p = .032), disappearing in a post-hoc comparison. These results resolve previous disparate studies to suggest that, although helmets can be uncomfortable, any effect of wearing a helmet on cognitive performance is at worst marginal.     

The relation between clothing thickness and cooling during motorcycling

The relation between clothing thickness and the cooling rate of a single motorcyclist has been determined at air temperatures between — 1°C and 24°C. The importance of a windproof oversuit and helmet liner have been studied. The effect of a fairing on the motorcycle has been measured. Much thicker and more windproof clothing than is generally available is needed to maintain normal body temperature on an unfaired motorcycle in winter: a mean clothing thickness of 20 mm inside a windproof oversuit is needed at 5°C, but is only effective if sweat does not accumulate inside the clothing.     

针对轻量化网络的安全帽检测方法

变电站内电气设备数量众多,在工人进行现场作业时需要对工人佩戴安全帽进行监测。由于机器学习的安全帽佩戴检测方法常常出现漏检和误检的情况,为提高对安全帽佩戴识别的准确率,同时加快识别速度,提出了一种基于YOLOv5s的轻量化卷积神经网络模型。通过引入RepVGG模块对网络主干进行轻量化,在网络后处理阶段通过Soft-NMS降低遮挡目标漏检率,以Mixup数据增强来扩充数据集,建立样本之间的线性关系,提升训练模型泛化性能,最后进行消融实验。实验结果表明,改进的模型的均值平均精度(mAP)达到80.4%,推理速度达到了83.3 f/s,为变电站安全帽佩戴检测提供了有效参考。     

基于改进YOLOv5的安全帽佩戴检测算法

针对现有安全帽佩戴检测干扰性强、检测精度低等问题,提出一种基于改进YOLOv5的安全帽检测新算法。首先,针对安全帽尺寸不一的问题,使用K-Means++算法重新设计先验框尺寸并将其匹配到相应的特征层;其次,在特征提取网络中引入多光谱通道注意力模块,使网络能够自主学习每个通道的权重,增强特征间的信息传播,从而加强网络对前景和背景的辨别能力;最后,在训练迭代过程中随机输入不同尺寸的图像,以此增强算法的泛化能力。实验结果表明,在自制安全帽佩戴检测数据集上,所提算法的均值平均精度（mAP）达到96.0%,而对佩戴安全帽的工人的平均精度（AP）达到96.7%,对未佩戴安全帽的工人的AP达到95.2%,相较于YOLOv5算法,该算法对佩戴安全帽的平均检测准确率提升了3.4个百分点,满足施工场景下安全帽佩戴检测的准确率要求。     

引入注意力机制的YOLOv5安全帽佩戴检测方法

对于钢铁制造业、煤矿行业及建筑行业等高危行业来说,施工过程中佩戴安全帽是避免受伤的有效途径之一.针对目前安全帽佩戴检测模型在复杂环境下对小目标和密集目标存在误检和漏检等问题,提出一种改进YOLOv5的目标检测方法来对安全帽的佩戴进行检测.在YOLOv5的主干网络中添加坐标注意力机制(coordinate attenti...     

Helmet design to facilitate thermoneutrality during forest harvesting

The purpose of this investigation was to evaluate the ventilatory characteristics of vented and non-vented helmets for use in forestry harvesting operations. A ventilation index developed by Birnbaum and Crockford (1978) was used to determine the ventilation capacity of twelve helmets varying in design and presence, location, and dimension of vents. Helmets with top vents had higher ventilation indices than non-vented, side and side/top-vented helmets. Ten physically fit men participated in a maximal oxygen consumption test and four trials wearing a non-vented, round-side-vented, round-top-vented or rectangulartop-vented helmet. Trials simulated typical summer environmental conditions (28°C, 80% rh) and physiological work loads (40% V O 2 max) experienced during forest harvesting in New Zealand. The temperature and humidity under the helmet was typically lowest when the helmet with the largest vented area (288 mm 2) in the crown was worn, although physiological responses (temperatures at the tympanum, ear and scalp; humidity at the centre of the helmet) of the participants to wearing the helmets differed significantly during some rest periods only. Top-vented helmets (in this experiment measured 288 mm 2) should be worn to minimize temperature and humidity of the head during forest harvesting operations.     

Helmet design to facilitate thermoneutrality during forest harvesting

The purpose of this investigation was to evaluate the ventilatory characteristics of vented and non-vented helmets for use in forestry harvesting operations. A ventilation index developed by Birnbaum and Crockford (1978) was used to determine the ventilation capacity of twelve helmets varying in design and presence, location, and dimension of vents. Helmets with top vents had higher ventilation indices than non-vented, side and side/top-vented helmets. Ten physically fit men participated in a maximal oxygen consumption test and four trials wearing a non-vented, round-side-vented, round-top-vented or rectangulartop-vented helmet. Trials simulated typical summer environmental conditions (28 degrees C, 80% rh) and physiological work loads (40% VO(2)max) experienced during forest harvesting in New Zealand. The temperature and humidity under the helmet was typically lowest when the helmet with the largest vented area (288 mm(2)) in the crown was worn, although physiological responses (temperatures at the tympanum, ear and scalp; humidity at the centre of the helmet) of the participants to wearing the helmets differed significantly during some rest periods only. Top-vented helmets (in this experiment measured 288 mm(2)) should be worn to minimize temperature and humidity of the head during forest harvesting operations.     

复杂姿态下的安全帽佩戴检测方法研究

在施工人员复杂姿态下,现有安全帽佩戴检测方法存在检测难度大,精度不高的问题,提出一种基于头部识别的安全帽佩戴检测方法。通过肤色特征识别和头部检测获取头部区域,并进行交叉验证确定头部区域,使用YOLOv4目标检测网络识别安全帽,通过安全帽区域与头部区域的位置关系判断安全帽佩戴情况。最后,通过实验对比分析其他安全帽佩戴检测方法的性能,对安全帽佩戴检测方法进行总结并提出展望。     

面向小目标的YOLOv5s安全帽佩戴检测

本文介绍了一种新的基于YOLOv5s的目标检测方法,旨在弥补当前主流检测方法在小目标安全帽佩戴检测方面的不足,提高检测精度和避免漏检.首先增加了一个小目标检测层,增加对小目标安全帽的检测精度;其次引入ShuffleAttention注意力机制,本文将ShuffleAttention的分组数由原来的64组减少为16组,更加有利于模型对深浅、大小特征的全局提取;最后增加SA-BiFPN网络结构,进行双向的多尺度特征融合,提取更加有效的特征信息.实验表明,和原YOLOv5s算法相比,改善后的算法平均精确率提升了1.7%,达到了92.5%,其中佩戴安全帽和未佩戴安全帽的平均精度分别提升了1.9%和1.4%.本文与其他目标检测算法进行对比测试,实验结果表明SAB-YOLOv5s算法模型仅比原始YOLOv5s算法模型增大了1.5M,小于其他算法模型,提高了目标检测的平均精度,减少了小目标检测中漏检、误检的情况,实现了准确且轻量级的安全帽佩戴检测.     

基于改进Faster RCNN的安全帽佩戴检测研究

针对已有安全帽佩戴检测算法对部分遮挡、尺寸不一和小目标存在检测难度大、准确率低的问题,提出了基于改进的Faster RCNN和多部件结合的安全帽佩戴检测方法。在原始Faster RCNN上运用多尺度训练和增加锚点数量增强网络检测不同尺寸目标的鲁棒性,并引入防止正负样本不均衡的在线困难样本挖掘策略,然后对检测出的佩戴安全帽工人和安全帽等采用多部件结合方法剔除误检目标。实验表明,相比于原始Faster RCNN,检测准确率提高了7%,对环境的适应性更强。     

改进YOLOv5s算法的安全帽佩戴检测

佩戴安全帽是施工过程中人员安全的重要保障之一,但现有的人工检测不仅耗时耗力而且无法做到实时监测,针对这一现象,提出了一种基于深度学习的安全帽佩戴检测算法。该算法以YOLOv5s网络为基础。在网络的主干网中引入CoordAtt坐标注意力机制模块,考虑全局信息,使得网络分配给安全帽更多的注意力,以此提升对小目标的检测能力;针对原主干网对特征融合不充分的问题,将主干网中的残差块替换成Res2NetBlock结构中的残差块,以此提升YOLOv5s在细粒度上的融合能力。实验结果表明：在自制的安全帽数据集中验证可知,与原有的YOLOv5算法相比,平均精度提升了2.3个百分点,速度提升了18 FPS,与YOLOv3算法相比,平均精度提升了13.8个百分点,速度提升了95 FPS,实现了更准确的轻量高效实时的安全帽佩戴检测。     

改进YOLOv3的安全帽佩戴检测方法

在生产和作业场地中，工人由于不佩戴安全帽而引发的安全事故时有发生。为了降低由于未佩戴安全帽而引发的安全事故发生率，提出了一种基于改进YOLO v3算法的安全帽佩戴检测方法。通过采用图像金字塔结构获取不同尺度的特征图，用于位置和类别预测；使用施工现场出入口监控视频作为数据集进行目标框维度聚类，确定目标框参数；在训练迭代过程中改变输入图像的尺寸，增加模型对尺度的适应性。理论分析和实验结果表明，在安全帽佩戴检测任务中，mAP（Mean Average Precision）达到了92.13%，检测速率提高到62?f/s，其检测准确率与检测速率相较于YOLO v3均略有提高，所提算法不仅满足安全帽佩戴检测中检测任务的实时性，同时具有较高的检测准确率。     

基于姿态估计的安全帽佩戴检测方法研究

针对现有安全帽佩戴检测方法在施工人员复杂姿态下检测难度大、精度不高的问题,提出一种基于姿态估计的安全帽佩戴检测方法。该方法在OpenPose姿态估计模型中引入残差网络优化特征提取,获得施工人员的骨骼点信息,并提出三点定位法,通过骨骼点位置信息确定头部区域以缩小检测范围。使用RetinaNet检测头部区域安全帽的佩戴情况,以解决安全帽与施工背景之间类极不平衡的问题。实验表明,该方法在检测精度上较其他方法有明显提高,并且对环境的适应性更强。     

CO detection in H2 reducing atmosphere with mini fuel cell

A prototype of a miniaturized fuel cell has been studied in order to detect carbon monoxide in hydrogen-rich atmosphere for PEMFC (proton exchange membrane fuel cell) applications. It consists on a single PEMFC (membrane-electrode-assembly supplied by CEA/LITEN) directly fed by the hydrogen-carbon monoxide mixture while the cathode is exposed to ambient air. Experiments have been carried out on a laboratory testing bench with simulated reforming gas. Two working modes have been studied. For low CO concentrations (≤20 ppm), the amperometric mode is suitable but a regeneration in air is necessary to obtain a good reversibility of the sensor response. On the contrary, for higher CO concentrations (250-4000 ppm), a good reversible response is observed without air regenerating by using a potentiometric or quasi-potentiometric mode. Therefore, this prototype of mini fuel cell sensor seems to be convenient for monitoring reformed gases either for low temperature PEMFC which are poisoned by very low traces of CO or for high temperature PEMFC which can operate at higher CO concentrations.     

基于改进区域卷积神经网络的安全帽佩戴检测

针对已有的安全帽佩戴检测算法对小尺寸目标和部分遮挡目标检测效果较差的问题,在区域卷积神经网络基础上,做出优化用于安全帽佩戴检测。在原始Faster RCNN的基础上使用多层卷积特征融合技术优化区域建议网络产生候选区域特征图,使用在线困难样本挖掘技术训练ROI网络,自动挑选出困难样本使训练更加有效。实验结果表明,相比原始的Faster RCNN算法,所提方法检测精度提高了4.73%,对部分遮挡和小尺寸目标均有较好的检测效果,对环境变化具有更强的适应性。     

Comparisons between Shikoro-type helmet with no hood and typical fire protective helmets with hood in a hot and humid environment

The purpose of this study was to evaluate physiological and subjective responses while wearing the Shikoro-type helmet for firefighters when compared to typical helmets. Eight firefighters conducted a 30-min exercise at a 5 km h^?1 in three helmet conditions at an air temperature of 32 °C with 70%RH. The results showed that no significant differences in rectal, mean skin temperature and physiological strain index among the three conditions were found during exercise and recovery. Skin temperatures on the cheek, ear and neck during exercise were significantly lower for the Shikoro-type condition (p < 0.05), but forehead temperature was greater for the Shikoro-type helmet when compared to the other conditions (p < 0.05). Statistical differences in thermal sensation and thermal comfort for overall and local body regions were not found among the three conditions. These results imply that the Shikoro-type helmet had local advantages in reducing skin temperatures on the face and neck.

Practitioner Summary: Firefighters wear their helmet with its hood to protect the head and neck but a Shikoro type helmet has no fire protective hood. This study aimed to evaluate the comfort function of Shikoro helmet along with typical helmets. The results demonstrated thermal benefits of the Shikoro helmet on the head.     

Effects of wearing aircrew protective clothing on physiological and cognitive responses under various ambient conditions

Heat stress can be a significant problem for pilots wearing protective clothing during flights, because they provide extra insulation which prevents evaporative heat loss. Heat stress can influence human cognitive activity, which might be critical in the flying situation, requiring efficient and error-free performance. This study investigated the effect of wearing protective clothing under various ambient conditions on physiological and cognitive performance. On several occasions, eight subjects were exposed for 3 h to three different environmental conditions; 0 degrees C at 80% RH, 23 degrees C at 63% RH and 40 degrees C at 19% RH. The subjects were equipped with thermistors, dressed as they normally do for flights (including helmet, two layers of underwear and an uninsulated survival suit). During three separate exposures the subjects carried out two cognitive performance tests (Vigilance test and DG test). Performance was scored as correct, incorrect, missed reaction and reaction time. Skin temperature, deep body temperature, heart rate, oxygen consumption, temperature and humidity inside the clothing, sweat loss, subjective sensation of temperature and thermal comfort were measured. Rises in rectal temperature, skin temperature, heart rate and body water loss indicated a high level of heat stress in the 40 degrees C ambient temperature condition in comparison with 0 degrees C and 23 degrees C. Performance of the DG test was unaffected by ambient temperature. However, the number of incorrect reactions in the Vigilance test was significantly higher at 40 degrees C than at 23 degrees C (p = 0.006) or 0 degrees C (p = 0.03). The effect on Vigilance performance correlated with changes in deep-body temperature, and this is in accordance with earlier studies that have demonstrated that cognitive performance is virtually unaffected unless environmental conditions are sufficient to change deep body temperature.