| Faster RCNN的交通场景下行人检测方法 |
| |
| 引用本文： | 徐向前,孙 涛.Faster RCNN的交通场景下行人检测方法[J].教育技术导刊,2020,19(4):67-70. |
| |
| 作者姓名： | 徐向前 孙 涛 |
| |
| 作者单位： | 上海理工大学 机械工程学院，上海 200093 |
| |
| 摘 要： | 为了提高行人检测的准确性和鲁棒性，针对现有行人检测方法准确率低且实时性不佳等问题，参考目标检测算法中快速区域卷积神经网络Faster RCNN算法，首先采用K-means聚类算法得到合适的宽高比，然后优化区域建议网络（RPN）结构，降低计算量，并通过比较MobileNet、VGG16、ResNet50特征提取网络效果优劣，提出改进Faster RCNN的交通场景下行人检测方法，在Caltech-NEW数据集上进行训练与测试。实验结果表明，该方法大幅提高交通场景下行人检测的实时性和准确性，在测试集上检测准确度达到87.5%，单张图片检测耗时为0.187s，相比现有其它方法，其检测效果更好。
|
| 关 键 词： | 卷积神经网络 行人检测 K-means算法 区域建议网络 |
| 收稿时间： | 2019-12-02 |
Pedestrian Detection Method in Traffic Scene Based on Faster RCNN |
| |
| XU Xiang-qian,SUN Tao.Pedestrian Detection Method in Traffic Scene Based on Faster RCNN[J].Introduction of Educational Technology,2020,19(4):67-70. |
| |
| Authors: | XU Xiang-qian SUN Tao |
| |
| Institution: | School of Mechanical Engineering， University of Shanghai for Science and Technology， Shanghai 200093，China |
| |
| Abstract: | In order to improve the accuracy and robustness of pedestrian detection， aiming at the problems of low accuracy and poor real-time performance of existing pedestrian detection methods， fast neural network of fast region convolution in target detection algorithm is referred. Firstly， K-means algorithm is used to obtain the appropriate aspect ratio， then the structure of region proposal network （RPN） is improved to reduce the amount of calculations. By comparing the performance of MobileNet， VGG16， and ResNet50 feature extraction networks， an improved detection method of traffic scene based on Faster RCNN is proposed， and the improved algorithm is trained and tested on the Caltech-NEW dataset. The experimental results show that the method greatly improves the real-time and accuracy of detection in traffic scenes. The mAP（mean Average Precision） on the test set is 87.5%， and the detection speed of a single picture is 0.187 seconds. Compared with other methods， this method is better. |
| |
| Keywords: | convolutional neural network pedestrian detection K-means algorithm RPN |
|
| 点击此处可从《教育技术导刊》浏览原始摘要信息 |
| 点击此处可从《教育技术导刊》下载免费的PDF全文 |
