采用改进的尺度不变特征转换及多视角模型对车型识别

Vehicle Recognition Using Improved SIFT and Multi-View Model

针对车型识别过程中车辆的姿态复杂以及采集图像时尺度缩放和光照等因素导致识别出现困难的问题,提出采用改进尺度不变特征转换(SIFT)及多视角的车型识别算法。该算法对尺度不变特征提取方法进行改进,并获取车型特征;通过视觉聚类对车辆进行多视角建模;利用最佳节点优先搜索算法完成特征向量的近邻搜索,并根据匹配相似度完成车型识别。实验结果表明,该算法所给出的车型识别方法具有可行性和有效性,可以在不同的图像畸变条件下保持稳定性,最终的车型识别效率也都可达到90%,所用时间要低于SIFT方法,处理时间在原SIFT方法的基础上降低了20.58%。

A method to recognize vehicles using an improved SIFT and multi-view model is proposed to improve the recognition problem caused by the complex posture of vehicle,scale zoom and illumination.The SIFT algorithm is improved to capture the feature of vehicles;the multi-view model of vehicles is built through visual clustering;The BBF algorithm is used to complete the nearest neighbor search of feature vectors,and vehicles are recognized by similarity matching.Experiments show that the proposed method of vehicle recognition is feasible and effective,and the method can keep stability in different conditions of image distortion.The rate of recognition can be up to 90%,the time is lower,and the processing time is reduced by 20.58% based on the original SIFT method.