基于改进无锚框网络的小尺度车辆目标检测方法

Vehicle recognition method based on improved anchor-free frame object detection network

小尺度车辆目标检测现已成为交通场景下目标检测中亟待解决的难题。对其中的难点进行研究,提出一种基于无锚框目标检测网络改进的算法。使用自适应特征提取方法,增强小尺度特征的表达,提高小尺度目标的特征提取能力;通过改进特征融合方法,将浅层信息逐层融合,解决特征丢失的问题。引入注意力增强方法,增加中心点预测能力,解决目标遮挡问题。实验结果表明,该算法在UA-DETRAC数据集上有很好的检测效果,较改进前车辆检测能力有较大提升,满足实时检测的要求,检测速度达到了59,平均精度均值为92.9%。

Small-scale vehicle target detection has now become an urgent problem to be solved in target detection in traffic scenes.The difficulties were studied,and an improved algorithm based on an anchorless frame target detection network was proposed.The adaptive feature extraction method was used to enhance the expression of small-scale features and improve the feature extraction ability of small-scale targets.By improving the feature fusion method,the shallow information was fused layer by layer to solve the problem of feature loss.The attention enhancement method was introduced to increase the center point prediction ability and solve the problem of target occlusion.Experimental results show that the improved method shows better detection effects on the UA-DETRAC data set compared with the original method,and meets the requirements of real-time detection.The detection speed reaches 59 fps,and the average accuracy is 92.9%.