基于改进YOLOv5的人员检测方法研究

为了解决在人员检测过程中，检测精度低、收敛速度慢问题，提出了一种基于改进的YOLOv5人员检测方法。首先，在主干网络(Backbone)和特征加强网络(Neck)中加入注意力机制(CBAM)，来解决Backbone特征提取能力不足问题，并加强Neck特征融合能力；然后，加入EIOU Loss，解决了计算宽高的差异值取代纵横比，同时引入Focal Loss解决难易样本不平衡问题，EIOU Loss在测试过程中，不仅仅加快模型的收敛速度，而且精度也有所提升。结果表明：在自制数据集和公开数据集CrowdHuman中，平均精度分别提高1.2%和1.6%，FPS每秒提升了11.91帧和6.44帧，漏检情况也有所降低。经过改进后的模型，实时性要求符合现实要求，更易于提取人员的特征信息，提升检测精度。

Research on human detection method based on improved YOLOv5

In order to solve the problem of low detection accuracy and slow convergence speed in the process of personnel detection, a YOLOv5 personnel detection method based on improvement is proposed. Firstly, the attention mechanism (CBAM) is added to the backbone network and the feature enhancement network (Neck) to solve the problem of insufficient feature extraction ability of Backbone and strengthen the ability of Neck feature fusion; Then, EIOU Loss was added to solve the problem of calculating the difference value of width and height instead of aspect ratio, and at the same time introduced Focal Loss to solve the problem of difficult sample imbalance, EIOU Loss not only accelerated the convergence speed of the model during the test, but also improved the accuracy. The results show that in the homemade dataset and the open dataset CrowdHuman, the average accuracy is improved by 1.2% and 1.6%, respectively, and the FPS is increased by 11.91 frames and 6.44 frames per second, and the missed detection situation is also reduced. After the improved model, the real-time requirements meet the realistic requirements, it is easier to extract the characteristic information of the personnel, and the detection accuracy is improved.