We publish our code and datasets for the paper "Traffic Intersection Re-Identification Using Monocular Camera Sensors". The code and datasets can be download by the related links!
-
python 3.6.8
-
PyTorch: Make sure to install the Pytorch version for Python 3.6 with CUDA support (code tested for CUDA 8.0, CUDA 9.0 and CUDA 10.0). I am using PyTorch 1.4.0.
-
Additional Python packages: numpy, matplotlib, Pillow, torchvision0.5.0
If you use our code or dataset, please consider referencing the following papers:
Xiong, L.; Deng, Z.; Huang, Y.; Du, W.; Zhao, X.; Lu, C.; Tian, W. Traffic Intersection Re-Identification Using Monocular Camera Sensors. Sensors 2020, 20(22), 6515.PDF
As the hubs of transportation networks, intersection is very valuable for research. To arrive at the destination, drivers usually construct the global driving route and especially preset the behaviors in intersections before moving the vehicle. All intersection images are saved in drivers' brain to form a topological map. Similar to human drivers, we argue that the required information in driving behavior decision for intelligent vehicles (IVs) such as fine intersection attributes and sparse positioning w.r.t. the intersection topological map can be achieved under a rational road intersection re-ID approach.
This project strives to explore intersection re-ID by monocular camera sensor, which strongly affects driving behavior decisions with given routes, yet has long been neglected by researchers. The visual sensor based intersection re-ID task in this project is defined as the multi-task including classification of intersection and its fine attributes, and the determination of global vehicle pose. For this project, we propose a Hybrid Double-Level re-identification approach which exploits two branches of Deep Convolutional Neural Network, and a mixed loss for network training. As no public datasets are available for the intersection re-ID task, we propose two intersection dataset named as "RobotCar Intersection" and "Campus Intersection".
Based on the prior work of RobotCar, we present "RobotCar Intersection" which covers 36588 images of eight intersections in different season and different time of the day.
-
Before using our dataset, the official Robotcar should be download by the link.
-
unzip all individual dataset into one folder named robotcar. Folder structure likes below.
|--robotcar
|--|--2014-05-19-13-20-57
|--|--|--image
|--|--|--|--0001.png, 0002.png, 0003.png, ....
|--|--2014-06-24-14-47-45
for training, data_train_previous60.txt
for testing, data_test_rest40.txt
the pretrained model can be download by this link Google Driver or Baidu Driver(提取码：5jsu).
This dataset consists of 3340 sperical panoramic images from eight intersections in the Jiading campus of Tongji University.
- download the dataset from online disk. The download link is Google Drive and Baidu Drive(提取码：br05)
- Folder structure likes below.
|--intersection_search_dwx
|--|--image
|--|--|--00001.jpeg, 00002.jpeg, 00003.jpeg, ....
for training, data_campus_train.txt
for testing, data_campus_test.txt
to identify new intersections, data_campus_new.txt
the pretrained model can be download by this link Google Driver or Baidu Driver(提取码：alcm).
We propose a Hybrid Double-Level (HDL) network for traffic intersection re-identification which is defined as the multi-task including classification of intersection and its fine attributes, and the determination of global vehicle pose.
We will compare our networks of different configuration with three baseline methods. The training detail and experimental result are shown in the paper.
txt file = (image path, intersection ID, Attribute ID, global ID)
data_campus_new.txt # the test images and labels of new 3 intersections.
data_campus_test.txt # the test images and labels of old 5 intersections.
data_campus_train.txt # the train images and labels of old 5 intersections.
data_test_rest40.txt # the test images and labels of RobotCar intersections.
data_train_previous60.txt # the train images and labels of RobotCar intersections.