摘要
传统车辆检测的算法无法自适应地完成复杂场景变化下目标特征提取,导致算法检测速度慢、检测效果差和检测精度低。提出了一种改进的Faster RCNN的车辆检测方法。这种方法设计了2个网络:一个准确的车辆候选区域检测网络(PVRNet)及车辆属性学习网络(VALNet)。通过大量的汽车图片数据样本,学习获得一个具有泛化能力强、检测以及定位准确率高的检测模型,其次将测试汽车图片输入该检测模型,得到可能的结果。可以处理复杂的视觉任务,避免人为设计车辆目标的特征,减少人员主观因素影响。实验结果表明,提出的方法显著提高了校园车辆检测性能,检测结果平均精度较高。
The traditional vehicle detection algorithm cannot adaptively complete the object features extraction under the complex scene,so the detection speed of the algorithm is slow,the detection effect is poor and the detection accuracy is low.An improved Faster RCNN method for vehicle detection is proposed.This method designs two networks:an accurate vehicle candidate area detection network(PVRNet)and a vehicle property learning network(VALNet).Through a large number of automobile image data samples,a detection model with strong generalization ability,high detection and positioning accuracy is learned.Then,the test samples are input into the detection model to get possible results.It can handle complex visual tasks,avoid the characteristics of artificial design of vehicle targets,and reduce the influence of subjective factors.Experimental results show that the proposed method significantly improves the detection performance of campus vehicles and the average accuracy of detection results is higher.
作者
李航
张琦
殷守林
孙可
LI Hang;ZHANG Qi;YIN Shoulin;SUN Ke(Software College, Shenyang Normal University, Shenyang 110034, China)
出处
《沈阳师范大学学报(自然科学版)》
CAS
2020年第1期77-84,共8页
Journal of Shenyang Normal University:Natural Science Edition
基金
辽宁省教育厅高等学校基本科研项目(LFW201702)。