基于 SLPP-SHOG 的红外图像车辆检测方法

摘要

针对红外图像的车辆检测，结合梯度方向直方图（HOG）特征与监督保局投影（SLPP），提出单帧图像车辆检测算法。首先，为增强特征描述能力、提高检测性能，在不增加特征维数的情况下，利用图像分割将区域的轮廓信息、灰度信息融入 HOG 特征中；其次，针对传统 HOG 特征维度过高，影响车辆检测效率以及准确率的问题，采用 SLPP 对提取的 SHOG特征进行降维；最后，利用极限学习机（ELM）对样本图像的低维特征进行训练得到 ELM分类器，实现车辆检测。本文以实拍红外图像作为实验数据，实验结果显示：针对红外图像的车辆检测，本文算法的检测性能较好，与原始 HOG 特征相比，本文所提 SLPP-SHOG 特征的特征维数由1764降至30，检测准确率升高16．03％，F1－measure 提高了8．79％，检测时间由5．7 ms降至2．6 ms。%For vehicle detection in infrared image,a vehicle detection algorithm based on histogram of oriented gradient (HOG)feature and supervised locality preserving projection (SLPP)was proposed.Firstly,in order to enhance the performance of detection,the grey information was obtained by using the image segmentation,and then the HOG fea-ture was obtained by enhancing the contour information without increasing the dimensionality of the feature.Secondly, the dimension of traditional HOG feature is too high which affect the efficiency and accuracy of vehicle detection.Therefore,the SLPP is used to reduce the dimension of features.Finally,to realize the vehicle detection,the extreme learning machine (ELM)is adopted to train the low dimensional feature of sample image.The presented ap-proach is tested in the reality infrared images.The experimental results show that the proposed method for vehicle de-tection in infrared images has better performance.Comparing with the result of HOG feature,the feature dimension of SLPP -SHOG decreases from 1 764 to 30;the detection precision increases by 1 6.03%;the F1 -measure rises by 8.79%;and the detection time reduces by 3.1 ms.