1)目标检测
介绍:给定图像,基于深度能算法得到图像中物体类别和位置,并以物体外接框的形式输出。已申请发明专利和软件著作权。
人脸检测
Demo: http://pan.baidu.com/s/1boWCkYN
车辆检测
Demo: http://pan.baidu.com/s/1eROy7tg
论文:Yan Tian, Huiyan Wang, Xun Wang, “Object Localization via Evaluation Multi-task Learning,” Neurocomputing, 253, 2017: 34-41. (SCI Q2, IF=4.43)
车道线检测
Demo: http://pan.baidu.com/s/1jI0kfgM
论文:
Yan Tian, Judith Gelernter, Xun Wang, Weigang Chen, Junxiang Gao, Yujie Zhang, and Xiaolan Li, “Lane Marking Detection via Deep Convolutional Neural Network,” Neurocomputing, 2018. (SCI Q2, IF=4.43)
专利:
田彦,王勋,王慧燕,华璟。基于集成学习级联分类器的车道线检测方法,授权号201610563188.X ,授权日2019-05-14。(可转让)
交通标识符检测(红色框)
论文:
Yan Tian, Xun Wang*, Judith Gelernter, Jianyuan Li, and Yizhou Yu, “Traffic Sign Detection And Recognition Using Recurrent Hourglass Network,”IEEE Transactions on Intelligent Transportations Systems, 2018. (SCI Q2, CCF-B, IF=6.31)
专利:
田彦,王勋,吴佳辰。一种基于多尺度循环注意力网络的交通标识符检测方法,授权号201810360701.4,授权日2020-08-14。(可转让)
2) 图像分割
介绍:提出一种自监督的图像分割方法,基于显著性自动分割出道路区域
效果:
论文:
Di Zhou, Yan Tian*, Wei-Gang Chen, Gang Huang. “Self-Supervised Saliency Estimation for Pixel Embedding in Road Detection”, IEEE Signal Processing Letters, 14(8), 2022: 1325-1329. (SCI Q2, IF=3.10)
专利:
徐照程, 田彦. 一种基于自监督学习显著性估计像素嵌入的道路检测方法. 申请号202110600086.1. 公开日20210812.
3)图像检索
介绍:给定一幅图像(没有检索关键词),基于深度学习算法从数据库中检索得到给定图像中物体的其它图像。
效果:
论文:
Yan Tian, Tao Chen, Guohua Cheng, Shihao Yu, Xi Li, Jianyuan Li, Bailin Yang*, “Global Context Assisted Structure-aware Vehicle Retrieval,” IEEE Trans on Intelligent Transportation Systems, 2020, in press. (CCF-B, IF=6.31)
专利:
钱小鸿,陈涛,李建元,田彦,虞世豪,一种卡口图像车辆检索方法及系统,授权号201811580165 ,申请日期2018-12-24.
4)流量预测
Traffic flow prediction plays an important role in intelligent transportation system. However, due to the fact that the traffic sensors are often manually controlled, the traffic flow data with varying length, irregular sampling and missing data is difficult to mine effectively. To deal with the missing value problem, we propose a new approach based on Long Short-Term Memory (LSTM) in this paper. Besides, multi-scale temporal smooth is employed to infer the lost data, and the prediction residual is learned by our approach. We demonstrate performance of our approach both on the PeMS dataset and our own traffic flow dataset. According to the experiments, our approach obtains better accuracy in traffic flow prediction compared with other approaches.
Results:
数据集:
https://www.kaggle.com/coplin/traffic/data
论文:
Yan Tian, Kaili Zhang, Jianyuan Li, Xianxuan Lin and Bailing Yang*, “LSTM Based Traffic Flow Prediction with Missing Data”, Neurocomputing, 2018. (SCI Q2, IF=4.43)
Bailing Yang, Shulin Sun, Kaili Zhang, Jianyuan Li and Yan Tian*, “Traffic Flow Prediction Using LSTM with Feature Enhancement, Neurocomputing, 2019. (SCI Q2, IF=4.43, ESI)
专利:
杨柏林,田彦,林贤煊,孙书林,张凯丽,基于平均偏移量平移的数据噪音点检测方法 ,授权号201711077817.9,授权日2020-10-16