密集人群检测是计算机视觉领域的重要任务,在监控、防踩踏、公共安全、交通监控等应用中具有广泛的应用价值。尽管基于深度学习的人群检测方法取得了显著进展,但在处理密集人群、小目标等复杂场景时存在不足,在涉及不同尺度的目标时无...密集人群检测是计算机视觉领域的重要任务,在监控、防踩踏、公共安全、交通监控等应用中具有广泛的应用价值。尽管基于深度学习的人群检测方法取得了显著进展,但在处理密集人群、小目标等复杂场景时存在不足,在涉及不同尺度的目标时无法准确地对其进行识别。本文提出了一种改进的YOLOv5密集人群检测算法,采用多尺度检测和图像分割技术提升算法在复杂场景下的检测能力。实验数据表明改进后的算法在密集多尺度场景下的效果优于传统的YOLOv5和YOLOv10,改进后的算法为密集人群检测任务提供了更全面有效的解决方案,有望在实际应用中更好地满足公共安全领域、交通领域等对密集人群检测的需求,对于预防踩踏等安全事故也具有重要意义。Dense crowd detection is an important task in the field of computer vision and has extensive application value in applications such as monitoring, anti-stampede, public safety, and traffic monitoring. Although crowd detection methods based on deep learning have made remarkable progress, there are still deficiencies when dealing with complex scenes such as dense crowds and small targets, and they cannot accurately identify targets of different scales. This paper proposes a dense crowd detection algorithm based on improved YOLOv5, which uses multi-scale detection and image segmentation technology to improve the detection ability of the algorithm in complex scenes. Experimental data shows that the improved algorithm is better than the traditional YOLOv5 and YOLOv10 in dense multi-scale scenes. The improved algorithm provides a more comprehensive and effective solution for dense crowd detection tasks, and is expected to better meet the needs of dense crowd detection in the fields of public safety and transportation in practical applications, and is also of great significance for preventing safety accidents such as stampedes.展开更多
文摘密集人群检测是计算机视觉领域的重要任务,在监控、防踩踏、公共安全、交通监控等应用中具有广泛的应用价值。尽管基于深度学习的人群检测方法取得了显著进展,但在处理密集人群、小目标等复杂场景时存在不足,在涉及不同尺度的目标时无法准确地对其进行识别。本文提出了一种改进的YOLOv5密集人群检测算法,采用多尺度检测和图像分割技术提升算法在复杂场景下的检测能力。实验数据表明改进后的算法在密集多尺度场景下的效果优于传统的YOLOv5和YOLOv10,改进后的算法为密集人群检测任务提供了更全面有效的解决方案,有望在实际应用中更好地满足公共安全领域、交通领域等对密集人群检测的需求,对于预防踩踏等安全事故也具有重要意义。Dense crowd detection is an important task in the field of computer vision and has extensive application value in applications such as monitoring, anti-stampede, public safety, and traffic monitoring. Although crowd detection methods based on deep learning have made remarkable progress, there are still deficiencies when dealing with complex scenes such as dense crowds and small targets, and they cannot accurately identify targets of different scales. This paper proposes a dense crowd detection algorithm based on improved YOLOv5, which uses multi-scale detection and image segmentation technology to improve the detection ability of the algorithm in complex scenes. Experimental data shows that the improved algorithm is better than the traditional YOLOv5 and YOLOv10 in dense multi-scale scenes. The improved algorithm provides a more comprehensive and effective solution for dense crowd detection tasks, and is expected to better meet the needs of dense crowd detection in the fields of public safety and transportation in practical applications, and is also of great significance for preventing safety accidents such as stampedes.
