|
|
|
|
|
|
|
|
|
|
|
Fig: Some of the Sample Video Clips of E-TUVD Dataset
|
|
|
|
|
The Extended Tripura University Video Dataset (E-TUVD) is being created in the Computer Vision Laboratory of Computer Science & Engineering Department, Tripura University (A Central University), India. Currently, E-TUVD contains 147 video clips depicting outdoor scenes with vehicles and pedestrians that have been captured in different real world scenarios and include a range of challenges under different complex atmospheric/ weather conditions. Each video clip has a duration of 2-5 minutes with a frame rate of 30 fps (Frame Per Second). |
|
| Importance/Utilities of the Dataset: |
Even though the previous datasets have advanced the research in moving object detection but these datasets are designed to maximally cover predefined fundamental challenges (e.g. complex background, shadow, occlusion, intermittent object motion, activity of objects, etc.) in moving object detection or tracking algorithms. Such intentionally selected videos may not be pragmatic in real world outdoor scenario and may over-fit the object detection models as the detection of moving objects in outdoor environment are mainly susceptible to atmospheric/ weather effects.
To further advance this area, our new designed video dataset provides a balanced coverage of real-world outdoor scenarios in degraded weather/ atmospheric conditions and contains ground truth images of salient moving objects that are unambiguously defined and annotated.
The major contributive features of this dataset are:
1. Background challenges:The video clips of E-TUVD depicts outdoor scenes either in static or dynamic background condition. For capturing the video with a static background, the camera is kept fixed with respect to the moving objects i.e. the background is static with respect to the moving objects. Conversely for dynamic background, the video is captured by mounting the camera on a moving vehicle (20∼30 km/h) where both the objects and background are moving simultaneously. Handling such background dynamics where the background is continuously moving with respect to the moving objects is a challenging task. The dataset contains 93 videos under static background condition and 54 videos under dynamic background condition.
2. Atmospheric/ Weather Challenges: The dataset includes urban scenes with buildings, trees, sky, vehicles and pedestrian with range from about 100 meters to about 3 KM. The dataset contains 22 videos under poor illumination condition, 34 videos in foggy weather condition, 12 videos in haze condition, 23 videos in dust condition and 27 videos in rainy condition as shown in Fig. 4. Moreover, the dataset also contain 29 video clips in clear day so as to facilitate the comparison of complex object detection models in both clear conditions and atmospheric degraded conditions.
3. Other Challenges: Beside these two major challenges, the dataset also contains scenes with multiple moving objects in a single frame, overlapping or occlusion between moving objects in a single frame, shadow and intermittent motions of objects. Although camouflage or poorly textured and camera jitter images are some of the considered key issues for moving object detection but we have not considered them in the categorical challenges of E-TUVD because almost all the real world video sequences contain some level of camouflage and camera jitter effect.
|
|
Dataset Statistics:
| Image Type |
Camera Model |
Background Condition |
Atmospheric/ Weather Conditions |
Total Videos |
| Foggy Condition |
Haze Condition |
Dust Condition |
Rain Condition |
Poor Illumination |
Clear Day |
|
| Visual |
Model: Nikon D5100
Lens: 18-55 mm
Shutter Speed: 1/125 - 1/200
Aperture: f/5.6 - f/8oc
Resolution: 1920 x 1080
Frame Rate: 30 FPS
|
Static |
24 |
07 |
14 |
15 |
15 |
18 |
93 |
| Dynamic |
10 |
05 |
09 |
12 |
07 |
11 |
54 |
|
| Total Number of Videos |
34 |
12 |
23 |
27 |
22 |
29 |
147 |
|
|
|
| Dataset Releasing Year: |
Dataset Still growing
|
|
| Size of Dataset: |
N/A
|
|
| Publications: |
-
Sourav Dey Roy and Mrinal Kanti Bhowmik "AWDMC-Net: Classification of Adversarial Weather Degraded Multiclass Scenes using a Convolution Neural Network", accepted in Computer Vision and Image Understanding, Elsevier, Indexed by Science Citation Index, Impact Factor: 3.876, Accepted on: 28th June 2022.
-
Sourav Dey Roy and Mrinal Kanti Bhowmik "Annotation and Benchmarking of a Video Dataset under Degraded Complex Atmospheric Conditions and Its Visibility Enhancement Analysis for Moving Object Detection", IEEE Transactions on Circuits and Systems for Video Technology, Indexed by Science Citation Index (SCI), Impact Factor: 4.133, (Accepted) on 23rd April 2020.
-
Sourav Dey Roy Mrinal Kanti Bhowmik, John Oakley, "A Ground Truth Annotated Video Dataset for Moving Object Detection in Degraded Atmospheric Outdoor Scenes", Proc. 2018 25th IEEE International Conference on Image Processing (ICIP), pp. 1318-1322. IEEE, 2018.
|
|
| Copyright: |
The E-TUVD dataset is the property of Tripura University. This data can only be used for research or academic purposes. Any commercial use of the data whatsoever or incorporation of the data into a larger database intended for public distribution must be done with the explicit written consent of the administrators.
|
|
| Created By: |
Research Group, Computer Vision Laboratory, Tripura University (A Central University).
|
|
| Accessed By: |
- Graduate School of Information Sciences, Tohoku University.
- Indian Institute of Information Technology, Design and Manufacturing, Kancheepuram, India.
- National Taiwan University, Taipei, Taiwan.
|
|
| |
For Access/Download E-TUVD Dataset Click Here
For Access/Download Train-Test Subset of E-TUVD Dataset Click Here
Click here to download User Agrement Form and send the user Agrement form to mrinalkantibhowmik@tripurauniv.ac.in, mkb_cse@yahoo.co.in and mkb.cse@gmail.com
|