基于CNN的交通标志识别方法研究

发布时间：2018-09-10 07:53

【摘要】：交通标志被广泛应用于道路交通中,由于受到许多不可控制因素如天气光照变化、物理遮挡、运动模糊等的影响,交通标志的准确检测和快速识别对研究者来说是一个重大的挑战。针对已有交通标志检测算法中存在误检较多的问题,本文提出一种结合布尔卷积神经网络的梯度方向直方图(HOG-BCNN,Histograms of Oriented Gradients-Boolean Convolutional Neural Network)与级联分类器的交通标志检测方法。该方法首先训练基于HOG特征的级联分类器,在含交通标志的道路场景图像提取潜在的候选区域,这些候选区域将作为建议窗口输入到一个特殊的CNN,它就像一个布尔逻辑器用以判断窗口是否为交通标志。布尔卷积神经网络模型作为级联分类器的最后一级对误检区域进行过滤,从而排除误检提高检测率。针对交通标志识别中网络结构复杂识别效率低的问题,本文提出了基于卷积神经网络的快速交通标识牌识别的方法(FTSR-CNN:Fast Traffic Sign Recognition Based on Convolution Neural Network)。该方法利用卷积核滑动滤波提取特征,通过池化技术降维,前向学习过程中求得网络的损失,反向传播中利用随机梯度下降法极小化损失,通过调整网络结构中的参数和激活函数类型来优化网络性能。实验结果表明,在交通标志的检测和识别方面,本文提出的方法能够实现较高的检测率和分类精度,同时保证检测和识别效率。最后,本文通过结合交通标志检测与识别的方法构建了交通标志识别系统。在行车记录仪中获取的多个视频上进行测试,取得了很好的实验效果。
[Abstract]:Traffic signs are widely used in road traffic, because of the influence of many uncontrollable factors, such as weather and light changes, physical occlusion, motion blur, etc. Accurate detection and rapid identification of traffic signs is a major challenge for researchers. In order to solve the problem of false detection in existing traffic sign detection algorithms, this paper presents a method of traffic sign detection based on gradient direction histogram (HOG-BCNN,Histograms of Oriented Gradients-Boolean Convolutional Neural Network) combined with Boolean convolution neural network (HOG-BCNN,Histograms of Oriented Gradients-Boolean Convolutional Neural Network) and cascaded classifier (Cascade classifier). The method first trains a cascade classifier based on HOG features to extract potential candidate regions from road scene images with traffic signs. These candidate regions will be input to a special CNN, as a suggested window, which is like a Boolean logic to determine whether the window is a traffic sign or not. The Boolean convolution neural network model is used as the last level of cascade classifier to filter the error detection area, thus eliminating the false detection and improving the detection rate. In view of the low efficiency of complex network structure recognition in traffic sign recognition, a fast traffic sign recognition method based on convolution neural network (FTSR-CNN:Fast Traffic Sign Recognition Based on Convolution Neural Network).) is proposed in this paper. In this method, the feature is extracted by convolution kernel sliding filter, the dimension is reduced by pool technique, the network loss is obtained in the process of forward learning, and the stochastic gradient descent method is used to minimize the loss in back propagation. The network performance is optimized by adjusting the parameters and activation function types in the network structure. The experimental results show that the method proposed in this paper can achieve high detection rate and classification accuracy, and ensure the efficiency of detection and recognition. Finally, a traffic sign recognition system is constructed by combining the detection and recognition of traffic signs. The test is carried out on several videos obtained from the CarLog, and the experimental results are very good.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】