多特征融合的车辆识别技术

Multi-feature fusion vehicle identification technology

针对运动车辆目标识别问题提出了一种自然场景下车辆识别方法。首先采用图像差分技术对目标车辆的显著特征进行统计学习,并将学习所得目标局部特征以及图像进行编码,根据以上两个信息实现目标车辆的显著性检测。其次针对车辆运动的复杂性,采用分块投影匹配方法进行全局运动估计和补偿,并利用差分技术进行运动特征检测。然后将目标车辆的显著性特征与运动特征进行融合,从而获得更精确的候选目标区域。最后对候选区域进一步使用视觉显著特征进行目标判别。实验表明该方法具有较好的目标判别性能,能较好地解决自然场景下运动车辆的识别问题。

A method of vehicle identification in natural scene was proposed for the target recognition of moving vehicles. Firstly, the image difference technique was used to study the significant characteristics of the target vehicle, and encode the local features and the image of the learning target, then the salient detection of the target vehicle was realized according to the above two information. Secondly, aiming at the complexity of vehicle movement, a block projection matching method was used for global motion estimation and compensation, and the difference technique was used to detect the motion feature. Then the target vehicle′ s saliency features and the motion features were fused to obtain more accurate candidate target areas. Finally, the candidate region was further used to discriminate the target by using the visual features. The experiment shows that the method has good performance of target discrimination and can solve the problem of vehicle motion recognition in natural scene.