Moving Object Detection in Degraded Atmospheric/ Weather Conditions

Dr. M.K. Bhowmik is actively involved in the domain of moving object detection in degraded atmospheric/ weather conditions. Detection of moving objects in dynamic variation of atmospheric/ weather is a challenging task because outdoor images often suffer from low contrast and limited visibility due to fog or other associate particles in the air that scatter the light in the atmosphere. Inclement weather has remained a challenge to many image processing applications as the performance of imaging devices reduce drastically under such conditions.



• Scope:


Generally North-Eastern (NE) state and other states of India shares international border and security plays a vital role. It has been observed that in an extreme atmospheric condition like fog, haze, dust, rain, etc., the outdoor scenes suffer from degradation and are not visible in naked eyes because of high loss in contrast. In such situation there is tendency of intruders to enter foreign lands for suspicious activities. So surveillance may plays an important role in change of illumination conditions to restrict the illegal threats of the state and for real time detection of suspicious activities. Modern and sophisticated surveillance systems with automatic detection installed in the border area may defend such activities.



• Our Contribution in the Proposed Domain:

Designing of Extended Tripura University Video Dataset (E-TUVD) in Degraded Atmospheric/ Weather Conditions:

In the last few decades, large datasets are designed to meet the increasing demands in developing and benchmarking new models for object detection. However, there is still lack of video datasets for designing comprehensive model for moving object detection that can provide a balanced coverage in weather/ atmosphere degraded outdoor scenes. Recognizing the importance of moving object detection to the computer vision and video processing communities, the research team have created a video dataset entitled as “Extended Tripura University Video Dataset (E-TUVD)” under different atmospheric/ weather conditions with unambiguously defined moving objects. Main contributive features of the designed Tripura University Video Dataset (E-TUVD) are:

1. The dataset comprises of video sequences of moving objects (especially of vehicles and pedestrians) in various weather/ atmospheric degraded challenges captured from different security and surveillance zones of Tripura.

2. The dataset contains 147 videos including 29 videos in clear condition, 22 videos in poor illumination condition, 34 videos in foggy condition, 12 videos in haze condition, 23 videos in dust condition and 27 videos in rainy condition captured under static and dynamic background conditions.

3. Along with the weather/ atmospheric challenges, the dataset also contains various representative challenges of moving object detection in real world i.e. intermittent object motion, camera jitter, shadows, overlapping of two moving objects, etc.

4. Each frame of TUVD contains multiple types of moving objects, e.g., pedestrians, various types of vehicles, bicyclists, motorbikes and pets. In addition, the captured scenes are mostly in urban areas, which correspond to larger surface variations due to the presence of objects such as trees, houses, warehouses, office buildings, streets, and residents.

5. For each video clip in the dataset, metrological data regarding weather information (hue, dew point, temperature, humidity, etc.) on the capturing day obtained from metrological department of Tripura State are also provided.

6. For all the captured video sequences, ground truth images of salient moving objects are also annotated in terms of object masks and human fixations (i.e. bounding box) to alleviate the ambiguity in defining and annotating salient moving objects and given along with this dataset. All these features of E-TUVD reflect its significance in the domain of moving object detection from outdoor scenes comprising of various atmospheric/ weather degraded challenges of real world scenarios. We expect this dataset will encourage the research community to make the strong assumptions of moving object detection benchmarks in adverse complex atmospheric/ weather conditions and to develop new methods that can be readily used in the many real time applications. Some of the sample images of the created dataset are highlighted in Figure.1.




Figure. 1. Some of the Sample Images of E-TUVD Dataset (a) Haze Condition; (b) Foggy Condition; (c)(d) Dust Condition; (e)(f) Rain Condition; (g)(h) Poor Illumination Condition; (i)(j)(k) Clear Day




• Featured Article(s) in the Proposed Domain:


1. 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), Volume: 31, Issue: 3, pp. 844-862, Impact Factor: 4.133, Accepted on: 29th April, 2020, DOI: 10.1109/TCSVT.2020.2991191.

2. Sourav Dey Roy Mrinal Kanti Bhowmik, John Oakley. "A Ground Truth Annotated Video Dataset for Moving Object Detection in Degraded Atmospheric Outdoor Scenes", Proceedings of 25th IEEE International Conference on Image Processing (ICIP) - Tier 2 Conference, Athens, Greece, pp. 1318-1322. IEEE, 2018, Electronic ISSN: 2381-8549.