• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.05a
• /
• pp.455-457
• /
• 2019

In this paper we study the effect of decomposed spatiotemporal convolutions for action recognition in videos. Our motivation emerges from the empirical observation that spatial convolution applied on solo frames of the video provide good performance in action recognition. In this research we empirically show the accuracy of factorized convolution on individual frames of video for action classification. We take 3D ResNet-18 as base line model for our experiment, factorize its 3D convolution to 2D (Spatial) and 1D (Temporal) convolution. We train the model from scratch using Kinetics video dataset. We then fine-tune the model on UCF-101 dataset and evaluate the performance. Our results show good accuracy similar to that of the state of the art algorithms on Kinetics and UCF-101 datasets.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2015.05a
• /
• pp.253-253
• /
• 2015

This research evaluated the effect of two main sources on the results of the ocean regional circulation model (ORCMs) during downscaling and nesting the results from the coarse data. The two sources should be the domain size, and temporal and spatial resolution different between driving and driven data. The Big-Brother Experiment is applied to examine the impact of them on the results of the ORCMs separately. Within resolution of 3km grid point ORCMs applying in the Big-Brother Experiment framework, it showed that the simulation results of the ORCMs depend on the domain size and specially the spatial and temporal resolution of lateral boundary conditions (LBCs). The domain size can be selected at 9.5 times larger than the interest area, and the spatial resolution between driving data and driven model can be up to 3 of ratio resolution and updating frequency of the LBCs can be up to every 6 hours per day.

• Journal of Electrical Engineering and Technology
• /
• v.10 no.4
• /
• pp.1851-1863
• /
• 2015

– In the paper, to handle the problem of traditional bag-of-features model ignoring the spatial relationship of local features in human action recognition, we proposed a Multiscale Spatial Position Coding under Locality Constraint method. Specifically, to describe this spatial relationship, we proposed a mixed feature combining motion feature and multi-spatial-scale configuration. To utilize temporal information between features, sub spatial-temporal-volumes are built. Next, the pooled features of sub-STVs are obtained via max-pooling method. In classification stage, the Locality-Constrained Group Sparse Representation is adopted to utilize the intrinsic group information of the sub-STV features. The experimental results on the KTH, Weizmann, and UCF sports datasets show that our action recognition system outperforms the classical local ST feature-based recognition systems published recently.

• Korean Journal of Remote Sensing
• /
• v.22 no.5
• /
• pp.337-350
• /
• 2006

The characteristics of neighborhood correlation images for change detection were explored at different spatial resolution scales. Bi-temporal QuickBird datasets of Las Vegas, NV were used for the high spatial resolution image analysis, while bi-temporal Landsat $TM/ETM^{+}$ datasets of Suwon, South Korea were used for the mid spatial resolution analysis. The neighborhood correlation images consisting of three variables (correlation, slope, and intercept) were evaluated and compared between the two scales for change detection. The neighborhood correlation images created using the Landsat datasets resulted in somewhat different patterns from those using the QuickBird high spatial resolution imagery due to several reasons such as the impact of mixed pixels. Then, automated binary change detection was also performed using the single and multiple neighborhood correlation image variables for both spatial resolution image scales.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2020.06a
• /
• pp.120-120
• /
• 2020

Spatial precipitation data is one of the essential components in modeling hydrological problems. The estimation of these data has achieved significant achievements own to the recent advances in remote sensing technology. However, there are still gaps between the satellite-derived rainfall data and observed data due to the significant dependence of rainfall on spatial and temporal characteristics. An effective approach based on the Convolutional Neural Network (CNN) model to correct the satellite-derived rainfall data is proposed in this study. The Mekong River basin, one of the largest river system in the world, was selected as a case study. The two gridded precipitation data sets with a spatial resolution of 0.25 degrees used in the CNN model are APHRODITE (Asian Precipitation - Highly-Resolved Observational Data Integration Towards Evaluation) and PERSIANN-CDR (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks). In particular, PERSIANN-CDR data is exploited as satellite-based precipitation data and APHRODITE data is considered as observed rainfall data. In addition to developing a CNN model to correct the satellite-based rain data, another statistical method based on standard deviations for precipitation bias correction was also mentioned in this study. Estimated results indicate that the CNN model illustrates better performance both in spatial and temporal correlation when compared to the standard deviation method. The finding of this study indicated that the CNN model could produce reliable estimates for the gridded precipitation bias correction problem.

• Proceedings of the KGS Conference
• /
• 2004.05a
• /
• pp.103-103
• /
• 2004

With the emergency of geographic information systems (GIS), the traditional layer based data model can only contain the spatial geometry and thematic attributes of phenomena. In real world, geographic phenomena have not only spatial geometry and thematic attribute but the temporal situation and the relationship between each phenomenon. (omitted)

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.11
• /
• pp.5459-5473
• /
• 2017

Recently, a tracking algorithm called the spatial-temporal context model has been proposed to locate a target by using the contextual information around the target. This model has achieved excellent results when the target undergoes slight occlusion and appearance changes. However, the target location in the current frame is based on the location in the previous frame, which will lead to failure in the presence of fast motion because of the lack of a prediction mechanism. In addition, the spatial context model is updated frame by frame, which will undoubtedly result in drift once the target is occluded continuously. This paper proposes two improvements to solve the above two problems: First, four possible positions of the target in the current frame are predicted based on the displacement between the previous two frames, and then, we calculate four confidence maps at these four positions; the target position is located at the position that corresponds to the maximum value. Second, we propose a target reliability criterion and design an adaptive threshold to regulate the updating speed of the model. Specifically, we stop updating the model when the reliability is lower than the threshold. Experimental results show that the proposed algorithm achieves better tracking results than traditional STC and other algorithms.

• Journal of Korean Society for Atmospheric Environment
• /
• v.16 no.4
• /
• pp.327-338
• /
• 2000

complexity in atmospheric environment coupled with shoreline and complex terrain often causes local variations of meteorology that are distinct from those representative over larger surrounding area, These kinds of local variations are less significant in usual long-term environmental impact analyses dealing with continuous plume. The variations could however be crucial in predicting dispersion of toxic substance released in a relatively small area for a short duration. In the present paper the effects of spatial and temporal resolution of diagnostic wind field on the dispersion of the released substance are investigated by using a puff model. A hypothetical release scenario assumes that a substance is released from a location in the Yochon Industrial Estate and passively dispersed within a few-kilometer distance for an hour. The results show that diagnostic analysis could resolve more spatial variations to some extent by employing smaller grid size. The peak concentrations and puff trajectories obtained from spatially -and/or tmeporally -varing diagnostic wind field are found appreciably different from those obtained from uniform wind field. Attention to high-resolution wind field in the both spatial and temporal spaces is called in the consequence analysis of toxic substance release.

• Proceedings of the KSRS Conference
• /
• v.1
• /
• pp.432-435
• /
• 2006

Potential application topics of KOMPSAT-3 image in the field of precision agriculture are suggested. The topics can be categorized as fundamental and applied ones that have contents of static and dynamic characteristics respectively. As fundamental topics, precision information of agriculture that is related to farmland and its crop attributes, precision information of rural infrastructure that is related to rural village and its facilities, precision information of stream environment that is related to rural water resources and its facilities, and precision information of eco-environment that is especially related to riparian ecology and environmental status are included. As applied topics, precision rural water resources that has thematic contents of continuous and event-based runoff, spatial and temporal soil moisture and evapotranspiration, precision agricultural watershed environment that has the contents of spatial and temporal soil loss, sediment and pollutants transport, and precision temporal and spatial crop growth that has the contents of temporal crop texture, spectral reflectance, leaf area index, spatial crop protein information.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.62 no.1
• /
• pp.51-60
• /
• 2020

The purpose of this study is to determine the spatial and temporal resolutions that can represent land surface characteristics comprised of various land use using Landsat/AMSR2-based soil moisture data. We estimated the Landsat (30 m×30 m)-based soil moisture values using the soil moisture regression model. Then, the Landsat (30 m×30 m)-based soil moisture (reference values) were resampled to the relatively coarse resolutions from 1 km to 4 km, respectively. Comparing the reference values to the resampled soil moisture values, we confirmed that uncertainties were increased with the spatial resolutions of 2 km~4 km indicating that the spatial resolution of 1 km×1 km is required to represent the complicated land surface. Also, the AMSR2 soil moisture values have less uncertainties compared to SMAP data with the temporal resolution of 1~2 days. Thus, our findings can be useful for various areas such as agriculture, hydrology, forest, etc.