基于特征融合和集成学习的隧道内车辆视频识别方法

A video vehicle recognition method based on feature fusion and ensemble learning

随着数字视频技术在交通领域的广泛应用,视频车辆识别成为了一个"ITS"领域的基础问题.视频图像中的车辆识别是一个典型的分类学习问题,针对这一问题,提出一种基于特征融合和集成机器学习的车辆识别算法.该方法首先获取视频序列中的兴趣区域;然后对兴趣区域提取纹理特征、Hu不变矩特征和小波特征,将这些特征组合成一种新的特征向量;然后将组合特征向量作为BP神经网络输入进行训练得到基分类器,最后利用Adaboost方法将BP神经网络集成得到强分类器.对所用方法进行了实验对比分析,其统计正检率、误检率、漏检率以及准确率等多参数结果均优于其他两种算法,实验验证该方法具有较好的识别率和鲁棒性.

With the wide application of digital video technology to the field of transportation,video vehicle recognition has become the basic problem in ＂ITS＂field.Vehicle recognition in video images is a typical classification learning problems.In order to solve this problem,a vehicle identification algorithm is proposed based on feature fusion and integrating machine learning.The method first gets the regions of interest in video sequences;then the texture features,Hu invariant moment features and wavelet features of region of interest are extracted;and these features are combined into a new feature vector.The combined feature vector is input into BP neural network for training under the Adaboost framework;and the base classifier is established.Finally,the strong classifier is obtained by integrating BP neural networks.A testing for the method is performed;the results show that the total positive rate,false positive rate,false negatives rate and accurate rate of the method are all better than those of the other two methods,so as to demonstrate that the presented method has better recognition rate and robustness.