Multi-task learning (MTL) has emerged as a crucial approach for addressing complex computer vision problems in autonomous driving, such as semantic segmentation, object detection, and monocular depth estimation. It reduces computational costs by enabling shared learning across interrelated tasks and gives a good understanding of the visual scene. Nonetheless, optimizing multi-task networks for varied tasks is a considerable problem owing to competing gradients and varying task-specific constraints. We propose in this study an attention-based technique to enhance multi-task network performance in autonomous driving. We dynamically control feature relevance across tasks by including attention mechanisms into convolutional neural networks (CNNs), therefore enabling the model to prioritize pertinent data for every task. Attention helps the model to concentrate on pertinent aspects for every activity, hence enhancing the alignment between the common representations and objectives tailored for each work. By allowing the model to acquire additional discriminative and task-specific features, integration of attention mechanisms helps to increase accuracy and robustness in demanding driving situations.

Monocular depth estimation is a critical component of autonomous driving, enabling vehicles to perceive and navigate their surroundings efficiently. Existing self-supervised methods often rely on complicated architectures and complex loss functions and face major challenges like inaccurate depth estimation for dynamic objects, blurred edge boundary, and edge flattening [1]. In this paper, we introduce a novel approach to leverage powers of high quality pseudo-depth supervision from DepthAnything-V2 model [2] to existing self-supervised approaches. We propose a novel dynamic mask generation method for any kind of dynamic objects including non-rigid objects in the scene, without complicated training procedures and loss functions. The simplicity of our approach, with competitive results compared to existing methods that rely on sophisticated techniques to solve basic problems, provides a robust and practical solution for real-world deployment. Extensive experiments on Cityscapes datasets verify the effectiveness of our method and establish it as a competitive alternative to existing monocular depth estimation models.

We propose a dense local region detector to extract features suitable for image matching and object recognition tasks. Whereas traditional local interest operators rely on repeatable structures that often cross object boundaries (e.g., corners, scale-space blobs), our sampling strategy is driven by segmentation, and thus preserves object boundaries and shape. At the same time, whereas existing region-based representations are sensitive to segmentation parameters and object deformations, our novel approach to robustly sample dense sites and determine their connectivity offers better repeatability. In extensive experiments, we find that the proposed region detector provides significantly better repeatability and localization accuracy for object matching compared to an array of existing feature detectors. In addition, we show our regions lead to excellent results on two benchmark tasks that require good feature matching: weakly supervised foreground discovery and nearest neighbor-based object recognition.

We introduce a fast deformable spatial pyramid (DSP) matching algorithm for computing dense pixel correspondences. Dense matching methods typically enforce both appearance agreement between matched pixels as well as geometric smoothness between neighboring pixels. Whereas the prevailing approaches operate at the pixel level, we propose a pyramid graph model that simultaneously regularizes match consistency at multiple spatial extents-ranging from an entire image, to coarse grid cells, to every single pixel. This novel regularization substantially improves pixel-level matching in the face of challenging image variations, while the "deformable" aspect of our model overcomes the strict rigidity of traditional spatial pyramids. Results on Label Me and Caltech show our approach outperforms state-of-the-art methods (SIFT Flow [15] and Patch-Match [2]), both in terms of accuracy and run time.

Designing a new sensor board is costly, especially for a production in a small quantity. By modularizing common functionalities, a large portion of the sensor board can be reused. In this work, we propose an Extension Board, a sensor board which is modularized into 3 parts. Power module, and MCU and RF module are shared, and only sensing module is redesigned for each sensor board. Diverse sensing modules are produced. The process was simple and inexpensive.

PEAKSAVE system is an energy monitoring service based on Wireless Sensor Networks (WSN). A smartphone is an important point of a system. Users can understand the energy consumption of each electric device and lighting in real time. Responsive energy monitoring service can help in reducing the waste of energy, especially in shaving electric load in a peak time.

We propose a dense local region detector to extract features suitable for image matching and object recognition tasks. Whereas traditional local interest operators rely on repeatable structures that often cross object boundaries (e.g., corners, scale-space blobs), our sampling strategy is driven by segmentation, and thus preserves object boundaries and shape. At the same time, whereas existing region-based representations are sensitive to segmentation parameters and object deformations, our novel approach to robustly sample dense sites and determine their connectivity offers better repeatability. In extensive experiments, we find that the proposed region detector provides significantly better repeatability and localization accuracy for object matching compared to an array of existing detectors. In addition, we show our regions lead to excellent results on two benchmark tasks that require good feature matching: weakly supervised foreground discovery, and nearest neighbor-based object recognition.

We present an approach to discover and segment foreground object(s) in video. Given an unannotated video sequence, the method first identifies object-like regions in any frame according to both static and dynamic cues. We then compute a series of binary partitions among those candidate “key-segments” to discover hypothesis groups with persistent appearance and motion. Finally, using each ranked hypothesis in turn, we estimate a pixel-level object labeling across all frames, where (a) the foreground likelihood depends on both the hypothesis's appearance as well as a novel localization prior based on partial shape matching, and (b) the background likelihood depends on cues pulled from the key-segments' (possibly diverse) surroundings observed across the sequence. Compared to existing methods, our approach automatically focuses on the persistent foreground regions of interest while resisting oversegmentation. We apply our method to challenging benchmark videos, and show competitive or better results than the state-of-the-art.

Micro Energy Efficiency System (MEES) provides energy saving while enabling each individual office in a large building to control heating, cooling, and electricity with its own policy. The usage of a decentralized independent control of each office, rather than a centralized one, is common in Korea. MEES provides measuring and controlling points at multiple granularities. An installation became easy and efficient using QR code, and the user configuration through an energy web portal further enhances the energy efficiency.

We present a feature matching algorithm that leverages bottom-up segmentation. Unlike conventional image-to-image or region-to-region matching algorithms, our method finds corresponding points in an “asymmetric” manner, matching features within each region of a segmented image to a second unsegmented image. We develop a dynamic programming solution to efficiently identify corresponding points for each region, so as to maximize both geometric consistency and appearance similarity. The final matching score between two images is determined by the union of corresponding points obtained from each region-to-image match. Our encoding for the geometric constraints makes the algorithm flexible when matching objects exhibiting non-rigid deformations or intra-class appearance variation. We demonstrate our image matching approach applied to object category recognition, and show on the Caltech-256 and 101 datasets that it outperforms existing image matching measures by 10~20% in nearest-neighbor recognition tests.

Evapotranspiration (ET) is significant and necessary to understand the hydrological cycle and to assess water resource at any region. The Moderate Resolution Imaging Spectroradiometer (MODIS) offers promising techniques to monitor regional or global ET patterns. Under cloudy conditions, however, some pixels contain missing data that hamper the continuous monitoring of ET. In this study, MODIS atmospheric and land products were used ET estimates under clear and partial clear sky condition, and atmospheric data produced by the Four-Dimensional Data Assimilation (FDDA) between MODIS products and the Fifth Generation Meso-scale Meteorological Model (MM5) and MODIS land products were used ET estimates under cloudy sky condition. MODIS ET under clear and cloudy sky conditions showed a good agreement with nine flux tower observations in Northeast Asia. These results indicate that MODIS can be applied to monitor ET with reasonable accuracy.

U-City is a city where diverse public information is provided through IT technology. In the past, IT infrastructure for public information was not considered in city planning. However, in recent construction of new cities, this kind of infrastructure is becoming necessary. Safety, transportation, and weather information are gathered and provided to residents through the Internet, mobile devices, etc [1]. To provide this kind of information, local governments operates U-City control center, and there is an issue of increased operating cost of the city. To solve this problem, U-City business platform is designed which can incorporate diverse commercial services. Different from public information platform, a private sector can participate and provide services. This work suggests Business Service Platform (BSP) system structure where USN-based services can be provided in U-City business platform. Then, USN-based service applications are introduced on U-City platform. Business Service Platform (BSP) is a service platform of U-City for services in a private sector, and provides overall functionalities required for the creation, distribution, and billing. A service provider for U-City can develop a new application using BSP. It can also register, distribute, and handle billing using functionalities of BSP. BSP started with a target advertisement service related to public transportation information as its initial service, however USN technology in diverse areas are expected to be applied to BSP in the future. To apply such diverse USN to BSP, a general framework should be provided to integrate USN, and a system is needed that each service provider can control. In this demo, we will explain U-City service platform, which will be deployed at CHEONGRA zone of Korea, and how it is integrated to USN to provide healthcare, smart grid services, and finally major system components.

We propose a space-time Markov random field (MRF) model to detect abnormal activities in video. The nodes in the MRF graph correspond to a grid of local regions in the video frames, and neighboring nodes in both space and time are associated with links. To learn normal patterns of activity at each local node, we capture the distribution of its typical optical flow with a mixture of probabilistic principal component analyzers. For any new optical flow patterns detected in incoming video clips, we use the learned model and MRF graph to compute a maximum a posteriori estimate of the degree of normality at each local node. Further, we show how to incrementally update the current model parameters as new video observations stream in, so that the model can efficiently adapt to visual context changes over a long period of time. Experimental results on surveillance videos show that our space-time MRF model robustly detects abnormal activities both in a local and global sense: not only does it accurately localize the atomic abnormal activities in a crowded video, but at the same time it captures the global-level abnormalities caused by irregular interactions between local activities.