基于多分支卷积神经网络的车辆图像比对方法

为了更有针对性地从车辆图像的不同区域提取出独特的图像特征,提出基于多分支卷积神经网络的车辆图像比对方法。首先,根据车牌定位结果获取比对车辆的车脸图像,并根据车脸图像的纹理丰富度将车脸图像划分为多个图像块;其次,使用多分支卷积神经网络分别提取各车脸图像块的深度特征;最后,通过计算车脸图像深度特征的相似度判定比对的车辆图像是否属于同种车型。实验表明,提出的方法能够提取有效的车辆图像各区域的深度特征,获得良好的车辆比对准确率,可用于套牌车辆识别。

面向视频侦查应用的跨摄像头车辆比对方法

针对视频侦查工作中跨摄像头实时精确查找犯罪嫌疑车辆的难题,提出了一种基于SIFT特征匹配算法的跨摄像头车辆快速自动比对方法:首先输入需要查找的监控视频重点时间段,然后利用帧差法从视频中提取包含嫌疑犯罪车辆的关键帧,使用二值化与腐蚀膨胀得到主体轮廓图像后提取SIFT特征点,最后对包含有嫌疑车辆的视频关键帧与案发时间段内记录了所有通过的车辆的视频关键帧进行自动的SIFT特征比对,其中特征点匹配数达到最小特征点匹配数的样本即可认定为所要追查的嫌疑车辆。实验结果表明,可以实现不同摄像头下的多品牌车辆的快速自动比对,能够进一步丰富视频侦查工作的技术手段。

Statistical Evaluation and Regression Analysis of Vehicle Sound Quality

A statistical approach to evaluate the subjective perception of the annoyance caused by the vehicle noise was presented in this paper. After recording the noises of Sanfeng, Huali and Xiali at speeds of 30, 40, 50, 60, 70 and 80 km/h respectively, the annoyance of the vehicle noises was evaluated in the testing room using paired comparison method, and the sound quality metrics and subjective annoyance were then distilled. Loudness, sharpness, roughness, periodicity and impulsiveness were selected for each of the vehicle noises. By correlation analysis method, it can be found that loudness has a higher correlation （0.91） with annoyance than other parameters. Meanwhile, sharpness, periodicity, roughness and impulsiveness have correlation with subjective perception with correlation coefficients being 0.84, -0.82, 0.62 and 0.87, respectively. The result of multiple regression analysis shows that calculated annoyance obtained by the regression equation can explain the perceptual annoyance and the regressed evaluation model is feasible to evaluate the sound quality of vehicle.

基于高清车牌识别的二义性路径识别系统

在高速公路路网结构复杂、多变的情况下,考虑到高速公路收费站的运营需求以及高清车牌识别技术的迅速崛起,文章提出了基于高清车牌识别的二义性路径识别系统,以解决现有路网条件下的二义性问题,实现通行费精确拆分,从而有效地提高了高速公路收费系统和相应服务器采集、传输、分析数据的精准性以及运营管理工作的效率。