摘要： 针对复杂背景下的目标车型识别问题，提出一种基于视频序列的检测识别算法。运用帧差序列图像进行背景建模与更新，采用背景差分和LBP纹理分析法进行运动车辆的分割及阴影消除，提出车辆形状投影量的概念，将视频车辆二维形状信息降至一维，并设计二维输入模糊分类器，根据形状投影量和车高/车长比，完成车型的多种类精细识别。实验结果验证了该算法的有效性。

关键词: 视频序列, 纹理分析, 车辆检测, 车型识别

Abstract: To solve the problem of vehicle type recognition, a novel algorithm is proposed. The background reconstruction is proposed based on difference frame sequence. The vehicle detection is finished based on background difference method and the shadow elimination is based on LBP texture analysis method. The concept of projection of vehicle shape is proposed, the 2D information of vehicle shape is reduced dimensionality and the robustness of algorithm is increased. A two-input fuzzy classifier is designed and the high-precision multi-class accurate vehicle recognition is finished by using the projection of vehicle shape and the ratio of height and length. Experimental results prove the validity and practicability of the algorithm.

Key words: video sequence, texture analysis, vehicle detection, vehicle type recognition

中图分类号: 

• TP311.52