摘要
针对某些简易快速车辆识别系统只能得到有限的车辆轮廓信息,增加了车型识别的难度的情况,提出一种基于混合轮廓特征的车型识别的新方法。首先根据车型识别系统获取尽可能精确的车辆轮廓,在经过简单的预分类之后分别提取出车型轮廓的七个轮廓矩不变量和有最强表征作用的Harris角点特征;进而进行特征级融合,将这两种特征融合构成混合特征;将融合后的新特征输入到RBF神经网络进行训练识别。实验结果表明在保证了较快识别速度的同时,有效地提高了识别率。
Aiming at the situation that some simple and fast vehicle recognition systems can only get limited information of vehicles' profiles and this increases the difficulty in vehicle recognition, we proPosed a new mixed contour feature-based vehicles type recognition method. First, we obtained vehicles profiles as accurate as possible according to vehicle type recognition system. After simply pre-classifying them we extracted 7 contour moment invariants and the Harris comer features with strongest characterisation role from vehicles profiles. And then we conducted the feature-level fusion to fuse and compose these two features to the mixed feature. We inputted the new fused feature into RBF neural network for recognition training, experimental result showed that this method improved the recognition rate effectively while ensuring the rapid recognition speed.
出处
《计算机应用与软件》
CSCD
2016年第2期142-145,149,共5页
Computer Applications and Software
基金
天津市科技支撑计划重点项目(10ZCKFSF01100)
天津市科技型中小企业创新基金项目(13ZXCXGX40400)