EyeonTraffic(EoT) Dataset

Dataset for analyzing lane-less traffic state behavior at intersections

EyeonTraffic (EoT) Dataset is the first aerial view data for defining Spatio-temporal annotations to estimate the traffic congestion state under lane-less behavior. There were a total of 3 intersections chosen for the EoT dataset with around 1 hour of aerial video recorded for each of the intersections, namely, Paldi (P), Nehru bridge - Ashramroad (N), and APMC market (A) in the city of Ahmedabad, India. These intersections were considered because of the diverse traffic conditions they present. While Paldi and Nehru bridge are four-way signalized intersections, the APMC market is a three-way non-signalized intersection. Hence, this dataset comprehensively covers a wide variety of traffic conditions for both signalized and non-signalized intersections. Details of the tracking annotations are found in SkyEye repository.

Paldi (P)	Nehru Bridge Ashram Road (N)
4-way signalized intersection	4-way signalized intersection
APMC market (A)
3-way unsignalized intersection

Spatio - Temporal Annotations

Spatial regions of traffic states for the above intersections are manually annotated as shown below

Red: Clump, Yellow: Neutral, Blue: Unclump

Paldi (P)	Nehru Bridge Ashram Road (N)
4-way signalized intersection	4-way signalized intersection
APMC market (A)
3-way unsignalized intersection

Spatial annotations can be downloaded here:Paldi, Nehru, APMC

Temporal annotations and corresponding videos can be downloaded using the below table

Paldi		Nehru		APMC
Video	Annotation	Video	Annotation	Video	Annotation
1_1.mp4	Paldi_1_1.csv	3_1.mp4	Nehru_3_1.csv	4_1.mp4	APMC_4_1.csv
1_2.mp4	Paldi_1_1.csv	3_2.mp4	Nehru_3_2.csv	4_2.mp4	APMC_4_2.csv
1_3.mp4	Paldi_1_1.csv	3_3.mp4	Nehru_3_3.csv	4_3.mp4	APMC_4_3.csv
NA	NA	3_4.mp4	Nehru_3_4.csv	NA	NA

Finally, the Spatio-Temporal annotations segmented at the rate 5fps combining spatial and temporal annotations and for 3 intersections can be downloaded from the below table:

State	Paldi	Nehru	APMC
Clump	P_C	N_C	A_C
Neutral	P_N	N_N	A_N
Unclump	P_U	N_U	A_U

The overall breakdown of EoT dataset is given below:

CNN Features

The tracks obtained for each of the Spatio-temporal regions are used to create a corresponding adjacency matrix based on the road user ids. The distance between two road users is converted into meters from pixel values. If the distance is less than μ= 10m, the corresponding entry is added to the adjacency matrix based on road width. The image representation of the adjacency matrices is sent as input to the VGG16 CNN architecture pre-trained on the ImageNet dataset. The input image is resized to 224×224 and a 147 dimension feature vector is extracted from the average pool layer.

The features can be downloaded here

License

This dataset is provided for academic and research purposes only.

Annotator

• K Naveen Kumar, PhD Research Scholar, Dept. of Computer Science and Engineering, Indian Institute of Technology Hyderabad, India

Citation

If you use this dataset, consider citing our paper.

@inproceedings{roy2020defining,
  title={Defining Traffic States using Spatio-temporal Traffic Graphs},
  author={Roy, Debaditya and Kumar, K Naveen and Mohan, C Krishna},
  booktitle={2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC)},
  pages={1--6},
  year={2020},
  organization={IEEE}
}

Acknowledgment

This work has been conducted as the part of SATREPS project M2Smart “Smart Cities development for EmergingCountries by Multimodal Transport System based on Sensing, Network and Big Data Analysis of Regional Transportation” (JPMJSA1606) funded by JST and JICA