基于TOPHAT-PCNN的图像车辆目标检测方法研究  

Research on vehicle detection based on TOPHAT-PCNN

作　　者：张栋冰[1] 

机构地区：[1]淮北师范大学计算机科学与技术学院,淮北235000

出　　处：《南京大学学报（自然科学版）》2017年第3期590-597,共8页Journal of Nanjing University（Natural Science）

基　　金：安徽省高等学校自然科学研究项目(KJ2014B23);安徽省级质量工程项目(2016jyxm0917)

摘　　要：对传统的车辆目标检测方法进行改进,提出了一种基于形态学高帽变换(TOPHAT)与脉冲耦合神经网络(PCNN)相结合的车辆目标检测方法.首先对交通图像进行形态学高帽变化提取图像的目标区域、然后分析了PCNN特征对车辆图像与非车辆图像的区分度,统计了熵特征和脉冲点火特征分别对原始图像和TOPHAT图像的有效性,选取了迭代平均熵作为车辆检测的有效特征,并采用滑窗的方式进行车辆检测,最后利用边缘密度信息对检测出的车辆目标进行后续验证.实验从有效性和准确性两方面进行验证,实验图片来自实际交通路口,结果表明:该方法能够有效地进行车辆目标检测,同时与其他车辆目标检测方法相比,具有检测率高、误检率低,消耗时间少等特点,能够较好的实现智能交通中车辆目标的快速检测.;On the basis of improving the performance of traditional vehicle targets detection method, this paper focuses on effectiveness and accuracy,and proposes a new method o~ vehicle targets detection based on morphology top-hat transformation combined with pulse coupled neural network （PCNN）. First, morphology top hat transformation is utilized for traffic image to extract the morphology edges of all objects. Then pulse coupled neural network（PCNN） is utilized to analyze the discrimination between vehicle morphology edge image and non-vehicle morphology edge image. In the process of analysis, we evaluate the effectiveness of entropy features （EF）and pulse firing characteristics（PFC）on both the original image and top-hat image, respectively. Iterative average entropy features （IAEF）is chosen as the best characteristics for vehicle targets detection according to a large number of image statistics,and top-hat image is more suitable for vehicle targets detection. We take advantage of iterative average entropy features（IAEF）to detect the vehicle targets through a sliding window,and the window size is adaptively de- termined according to the test images. Finally,the edge density information（EDI）is utilized to confirm the detected vehicle targets,only high edge density information images can be recognized as a vehicle image. The experiments in this paper are verified from two aspects of validity and accuracy, and all test images are come from different actual traffic junctions in Wuhan. The experiments show that the method proposed in this paper can effectively for vehicle targets detection. Compared with other vehicle target detection methods, our method has higher detection rate and lower {alse detection rate and less time, and can better realize the goal of rapid vehicle targets detection in the intelligent transportation system（ITS）with brilliant prospects in applications.

关 键 词：高帽变换 脉冲耦合神经网络 车辆检测 迭代平均熵 边缘密度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]