摘要
对行人和车辆的识别是行车记录仪识别系统的重要组成部分,为满足车载识别系统对算法模型检测实时性和简洁性的需求,以传统Mobilenet为基础网络,提出了一种车载图像识别改进算法。该算法优化了Mobilenet网络的Con DW3×3、Conv3×3卷积层相关参数,网络的计算量明显减少,识别速度有效提高。利用行车记录仪获得的实际数据集进行训练,实验结果表明,改进算法在识别速率上提高了6.6%。相比同类网络模型Squeezenet,训练的迭代次数显著降低。
The recognition of pedestrians and vehicles is an important part of the recognition system of dashcam.In order to meet the requirements of the on-board recognition system for the real-time and simplicity of the algorithm model detection,an improved algorithm for on-board image recognition is proposed based on the traditional Mobilenet network.This algorithm optimizes Con DW3×3 and Conv3×3 convolutional layer related parameters of Mobilenet network,which significantly reduces the computational amount of network and effectively improves the recognition speed.The actual data set obtained from the dashcam was used for training.Experimental results show that the improved algorithm improves the recognition rate by 6.6%.Compared with the similar network model Squeezenet,training iterations were significantly reduced.
作者
张凯斐
王翠娥
ZHANG Kai-fei;WANG Cui-e(Department of Computer Science and Technology,Lüliang University,Lishi Shanxi 033001,China)
出处
《吕梁学院学报》
2021年第2期15-20,共6页
Journal of Lyuiang University
基金
吕梁学院教学改革创新项目(JYZD201703)。
