基于增强聚合通道特征的实时行人重识别
发布时间：2018-11-15 08:07
【摘要】：由于目标姿态、摄像头角度、光线条件等因素的影响,行人重识别仍然是一个具有挑战性的问题。目前大多数方法主要注重提高重识别精度,对实时性考虑较少。因此,本文提出了一种基于增强聚合通道特征(ACF)的实时行人重识别算法。利用ACF对行人进行检测,并在此基础上,结合直方图特征和纹理特征构成增强ACF,作为行人重识别的特征描述子。利用测度学习方法对重识别模型进行训练。在4个数据集上的实验结果表明,与传统的重识别特征相比,提出的特征描述子逼近最好的重识别准确率,并且具有更快的计算速度。整个行人检测与重识别系统的运行速度达到10 frame·s~(-1)以上,基本可以满足实时行人重识别的需求。
[Abstract]:Due to the influence of target attitude, camera angle, light condition and so on, pedestrian recognition is still a challenging problem. At present, most of the methods mainly focus on improving the recognition accuracy, but less on the real-time. Therefore, a real-time pedestrian recognition algorithm based on enhanced aggregate channel feature (ACF) is proposed in this paper. Using ACF to detect pedestrians and combining histogram features and texture features, an enhanced ACF, is used as a feature descriptor for pedestrian recognition. The method of measure learning is used to train the recognition model. The experimental results on four datasets show that the proposed feature descriptor approximates the best recognition accuracy and has a faster computing speed than the traditional re-recognition features. The speed of the whole pedestrian detection and recognition system is more than 10 frame s ~ (-1), which can meet the requirement of real-time pedestrian recognition.
【作者单位】： 中国人民解放军空军航空大学飞行器控制系;
【基金】：国家自然科学基金(6160011396) 吉林省教育厅“十三五”科学技术研究项目(吉教科合字[2016]第515号)
【分类号】：TP391.41
，
本文编号：2332692
[Abstract]:Due to the influence of target attitude, camera angle, light condition and so on, pedestrian recognition is still a challenging problem. At present, most of the methods mainly focus on improving the recognition accuracy, but less on the real-time. Therefore, a real-time pedestrian recognition algorithm based on enhanced aggregate channel feature (ACF) is proposed in this paper. Using ACF to detect pedestrians and combining histogram features and texture features, an enhanced ACF, is used as a feature descriptor for pedestrian recognition. The method of measure learning is used to train the recognition model. The experimental results on four datasets show that the proposed feature descriptor approximates the best recognition accuracy and has a faster computing speed than the traditional re-recognition features. The speed of the whole pedestrian detection and recognition system is more than 10 frame s ~ (-1), which can meet the requirement of real-time pedestrian recognition.
【作者单位】： 中国人民解放军空军航空大学飞行器控制系;
【基金】：国家自然科学基金(6160011396) 吉林省教育厅“十三五”科学技术研究项目(吉教科合字[2016]第515号)
【分类号】：TP391.41
，
本文编号：2332692
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2332692.html
