摘要
基于提高红外图像行人检测准确率的目的,提出了一种基于多特征的红外行人检测算法。首先提取训练样本的梯度方向直方图特征和强度自相似性特征,利用二者相结合得到联合特征训练支持向量机(SVM),之后利用滑动窗口法対整幅红外图像进行遍历,用训练好的SVM进行分类检测。在LSI Far Infrared Pedestrian Dataset数据库上实验证明,基于多特征的检测方法相较于单一特征的方法提高了红外行人检测的精度,降低了误检率和漏检率。
In order to improve the accuracy of pedestrian detection in infrared images, an infrared pedestrian detection method is proposed in this paper. Firstly extract train samples＇ Histogram of Oriented Gradients feature and Intensity Self-Similarity feature, combine these two features to train support vector machine （SVM）, then use sliding window method to traverse an infrared image, the trained SVM is used to classification and detection. Experiments in LSI Far Infrared Pedestrian Dataset prove that based on multi-features method compared with based on single feature method improve infrared pedestrian detection accuracy, reduce the false positive rate and miss rate.
出处
《电子设计工程》
2016年第4期182-185,189,共5页
Electronic Design Engineering
关键词
红外行人检测
梯度方向直方图
强度自相似特征
支持向量机
infrared pedestrian detection
histogram of Oriented Gradients
Intensity Self-Similarity
Support Vector Machine
