基于视觉图像与激光点云融合的交通标志快速识别方法

Fast recognition method of traffic signs based on fusion of visual image and laser point cloud

摘要:

交通标志对车辆交通起到重要作用和意义,而智能交通中交通标志识别由于标志特征提取效果差,导致识别率低、识别时间长,因此,提出一种新的基于视觉图像与激光点云融合的交通标志快速识别方法。采用双边滤波方法预处理原始激光点云数据；通过归一化处理得到视觉图像激光点云融合的目标空间激光点云位置测距数值。通过测距值获取目标图像位置,归一化处理交通标志视觉图像,引入k均值聚类算法(k-means clustering algorithm)二聚类处理图像,采用制作的切割模板切割图像感兴趣区域(ROI),提取交通标志图像的深度特征,结合卷积神经网络二次过滤特征,重新标定二次过滤后的特征,最终利用卷积神经网络模型实现交通标志快速识别。经实验对比证明,采用所提方法的提取各个类型交通标志特征的提取效果较好,并且识别率达到89.74%,识别时间仅为13.1s,干扰下识别时间最高仅为15.1s,验证了该方法可以快速且准确识别各个类型的交通标志。

Abstract:

Traffic signs play an important role and significance in vehicle traffic. However, in intelligent traffic, the recognition rate of traffic signs is low and the recognition time is long due to the poor feature extraction effect. Therefore, a new rapid recognition method of traffic signs based on visual image and laser point cloud fusion is proposed. Two-sided filtering method was used to preprocess the original laser point cloud data. The location ranging values of laser point cloud in target space are obtained by normalized processing. The location of the target image is obtained by the ranging value, the traffic sign visual image is processed in a normalized way, the k-means clustering algorithm is introduced to process the image by diclustering, the ROI of the image is cut by the cutting template made, and the depth features of the traffic sign image are extracted. Combined with the secondary filtering features of convolutional neural network, the features after secondary filtering are re-calibrated, and finally the convolutional neural network model is used to realize the rapid recognition of traffic signs. The experimental comparison shows that the proposed method has a good feature extraction effect for all types of traffic signs, and the recognition rate reaches 89.74%, the recognition time is only 13.1s, and the highest recognition time is only 15.1s under interference, which verifies that this method can quickly and accurately identify all types of traffic signs.