由于现有的城市交通汽车目标检测方法主要使用神经网络来定位回归检测区域,这种方法忽略了特征在损失函数中的关联性,易受到场景立体匹配距离变化的影响,进而导致平均检测精度较低.基于级联卷积神经网络,设计一种全新的城市交通汽车目标图像检测方法.提取城市交通汽车目标检测特征,采用逐级匹配法提升检测样本质量.不同类型的检测目标的IOU分配阈值不同,利用Iterative Bbox at Inference级联卷积神经网络进行分类回归处理,得到基于级联神经网络的汽车目标检测损失函数,对于每个栅格,需要预设先验框根据损失函数,计算预测参数,设计城市交通汽车多目标检测算法,从而实现城市交通汽车目标检测.实验结果表明:该设计方法的平均检测精度较高,说明所设计方法的检测效果较好,具有较高的准确性,有一定的应用价值,能够为城市交通安全性的提升作出一定的贡献.
Target Image Detection Method of Urban Transportation Vehicle Based on Cascade Convolutional Neural Network
due to the existing urban traffic car target detection method mainly use neural network to locate the regres-sion detection area,this method ignores the characteristic correlation in the loss function,vulnerable to the influence of scene stereo matching distance change,leading to the average detection accuracy is low,so this paper is based on the cascade convolutional neural network,design a new urban traffic car target image detection method.The target detection characteristics of urban transportation vehicles are extracted,and the step matching method was adopted to improve the quality of detection samples.Different types of detection target IOU distribution threshold is different,using Iterative Bbox at inference cascade convolutional neural network classification regression processing,get car target detection loss function based on the cascade neural network,for each grid,need preset prior box according to the loss function,calculation pre-diction parameters,and the design of urban traffic car multi-target detection algorithm,so as to realize the urban traffic car target detection.The experimental results show that the average detection accuracy of the design method is high,which shows that the design method has good detection effect,high accuracy and certain application value.
cascade convolutional neural networkurban trafficautomobile targetimage detectionloss function
万方数据
维普
