针对一般手势识别算法的参数量、计算量和精度难以平衡的问题,提出一种轻量化篮球裁判手势识别算法。该算法在YOLOV5s(You Only Look Once Version 5s)算法的基础上进行重构:首先,用Involution算子替代CSP1_1的卷积算子,以扩大上下文信...针对一般手势识别算法的参数量、计算量和精度难以平衡的问题,提出一种轻量化篮球裁判手势识别算法。该算法在YOLOV5s(You Only Look Once Version 5s)算法的基础上进行重构:首先,用Involution算子替代CSP1_1的卷积算子,以扩大上下文信息捕获范围并减少核冗余;其次,在C3模块后加入协同注意力(CA)机制,以得到更强的手势特征提取能力;然后,用轻量化内容感知上采样算子改进原始上采样模块,并将采样点集中在目标区域而忽略背景部分;最后,利用以SiLU作为激活函数的Ghost-Net进行轻量化剪枝。在自制的篮球裁判手势数据集上的实验结果表明,该轻量化篮球裁判手势识别算法的计算量、参数量和模型大小分别为3.3 GFLOPs、4.0×10^(6)和8.5 MB,与YOLOV5s算法相比,分别减少了79%、44%和40%,mAP@0.5为91.7%,在分辨率为1920×1280的比赛视频上的检测帧率达到89.3 frame/s,证明该算法能满足低误差、高帧率和轻量化的要求。展开更多
为了方便观众更好地在观看比赛直播和录像时理解裁判手势的含义,或帮助录像分析师分析比赛视频,设计了一种实时篮球裁判手势检测与识别系统Yolov5-BR(Yolov5-Basketball Referee).首先,采用目标检测中的Yolov5算法为基础模型,对其边界...为了方便观众更好地在观看比赛直播和录像时理解裁判手势的含义,或帮助录像分析师分析比赛视频,设计了一种实时篮球裁判手势检测与识别系统Yolov5-BR(Yolov5-Basketball Referee).首先,采用目标检测中的Yolov5算法为基础模型,对其边界框的交并比(intersection over union,IoU)损失函数完全交并比(complete intersection over union,CIoU)进行加权处理,增强预测框的鲁棒性;其次,在C3模块后加入注意力机制,产生更具分辨性的特征表示,从而提升网络识别性能;此外,在检测层头部融入自适应特征融合机制,充分利用图像高层语义信息;最后,对目标置信度损失函数进行不对等加权处理,从而提高对小目标检测的鲁棒性.在自制的裁判手势数据集上,Yolov5-BR取得了95.4%的mAP值,本地视频检测速率为55.5帧/s,外接摄像头分辨率为1280×960,检测速率为25帧/s.实验结果表明,Yolov5-BR相对于原始模型在检测裁判手势的性能上有所提升,保持了较高的准确率、稳定性与实时性.展开更多
In the context that global shipbuilding market in 2003 continued to climb, shipyards in Japan and South Korea,the two largest world shipbuilding nations today, achieved a better performance in 2003 than in 2002. The t...In the context that global shipbuilding market in 2003 continued to climb, shipyards in Japan and South Korea,the two largest world shipbuilding nations today, achieved a better performance in 2003 than in 2002. The twocountries have set out and implemented a number of industry policies and measures based on the development ofthe world shipbuilding industry and the current status of shipyards in both countries. In addition, both nations haveattached more importance to investment in technologies and development of new products. On the other hand,shipyards in Europe are still in difficulties and many shipbuilding firms have already shifted their focus onto militaryships and high value-added ships. For this, European countries have also set out policies to subsidize the industry.展开更多
文摘针对一般手势识别算法的参数量、计算量和精度难以平衡的问题,提出一种轻量化篮球裁判手势识别算法。该算法在YOLOV5s(You Only Look Once Version 5s)算法的基础上进行重构:首先,用Involution算子替代CSP1_1的卷积算子,以扩大上下文信息捕获范围并减少核冗余;其次,在C3模块后加入协同注意力(CA)机制,以得到更强的手势特征提取能力;然后,用轻量化内容感知上采样算子改进原始上采样模块,并将采样点集中在目标区域而忽略背景部分;最后,利用以SiLU作为激活函数的Ghost-Net进行轻量化剪枝。在自制的篮球裁判手势数据集上的实验结果表明,该轻量化篮球裁判手势识别算法的计算量、参数量和模型大小分别为3.3 GFLOPs、4.0×10^(6)和8.5 MB,与YOLOV5s算法相比,分别减少了79%、44%和40%,mAP@0.5为91.7%,在分辨率为1920×1280的比赛视频上的检测帧率达到89.3 frame/s,证明该算法能满足低误差、高帧率和轻量化的要求。
文摘为了方便观众更好地在观看比赛直播和录像时理解裁判手势的含义,或帮助录像分析师分析比赛视频,设计了一种实时篮球裁判手势检测与识别系统Yolov5-BR(Yolov5-Basketball Referee).首先,采用目标检测中的Yolov5算法为基础模型,对其边界框的交并比(intersection over union,IoU)损失函数完全交并比(complete intersection over union,CIoU)进行加权处理,增强预测框的鲁棒性;其次,在C3模块后加入注意力机制,产生更具分辨性的特征表示,从而提升网络识别性能;此外,在检测层头部融入自适应特征融合机制,充分利用图像高层语义信息;最后,对目标置信度损失函数进行不对等加权处理,从而提高对小目标检测的鲁棒性.在自制的裁判手势数据集上,Yolov5-BR取得了95.4%的mAP值,本地视频检测速率为55.5帧/s,外接摄像头分辨率为1280×960,检测速率为25帧/s.实验结果表明,Yolov5-BR相对于原始模型在检测裁判手势的性能上有所提升,保持了较高的准确率、稳定性与实时性.
文摘In the context that global shipbuilding market in 2003 continued to climb, shipyards in Japan and South Korea,the two largest world shipbuilding nations today, achieved a better performance in 2003 than in 2002. The twocountries have set out and implemented a number of industry policies and measures based on the development ofthe world shipbuilding industry and the current status of shipyards in both countries. In addition, both nations haveattached more importance to investment in technologies and development of new products. On the other hand,shipyards in Europe are still in difficulties and many shipbuilding firms have already shifted their focus onto militaryships and high value-added ships. For this, European countries have also set out policies to subsidize the industry.