• Journal of Information Processing Systems
• /
• v.19 no.4
• /
• pp.439-449
• /
• 2023

This paper deals with a resource efficient artificial intelligence (AI) service architecture for multi-channel video streams. As an AI service, we consider the object detection model, which is the most representative for video applications. Since most object detection models are basically designed for a single channel video stream, the utilization of the additional resource for multi-channel video stream processing is inevitable. Therefore, we propose a resource efficient AI service framework, which can be associated with various AI service models. Our framework is designed based on the modular architecture, which consists of adaptive frame control (AFC) Manager, multiplexer (MUX), adaptive channel selector (ACS), and YOLO interface units. In order to run only a single YOLO process without regard to the number of channels, we propose a novel approach efficiently dealing with multi-channel input streams. Through the experiment, it is shown that the framework is capable of performing object detection service with minimum resource utilization even in the circumstance of multi-channel streams. In addition, each service can be guaranteed within a deadline.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.12
• /
• pp.4776-4794
• /
• 2020

We propose a rapid adaptive learning framework for streaming object detection, called EER-ASSL. The method combines the expected error reduction (EER) dependent rollback learning and the active semi-supervised learning (ASSL) for a rapid adaptive CNN detector. Most CNN object detectors are built on the assumption of static data distribution. However, images are often noisy and biased, and the data distribution is imbalanced in a real world environment. The proposed method consists of collaborative sampling and EER-ASSL. The EER-ASSL utilizes the active learning (AL) and rollback based semi-supervised learning (SSL). The AL allows us to select more informative and representative samples measuring uncertainty and diversity. The SSL divides the selected streaming image samples into the bins and each bin repeatedly transfers the discriminative knowledge of the EER and CNN models to the next bin until convergence and incorporation with the EER rollback learning algorithm is achieved. The EER models provide a rapid short-term myopic adaptation and the CNN models an incremental long-term performance improvement. EER-ASSL can overcome noisy and biased labels in varying data distribution. Extensive experiments shows that EER-ASSL obtained 70.9 mAP compared to state-of-the-art technology such as Faster RCNN, SSD300, and YOLOv2.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.23 no.56
• /
• pp.19-34
• /
• 2000

Impact of business process re-engineering requires the fundamental rethinking of how information systems are analyzed and designed. It is no longer sufficient to establish a monolithic system for fixed business environments. Information systems must be adaptive in nature. This demand is also applied in production domain. Enabling concept for the adaptive information system is reusability. This paper presents a new object-oriented analysis process for creating such reusable software components in production domain, especially for production planning and scheduling. Our process called MeCOMA is based on three meta-models: physical object meta-model, data object meta-model, and activity object meta-model. After the three meta-models are extended independently for a given production system, they are collaboratively integrated on the basis of integration pattern. The main advantages of MeCOMA are (1) to reduce software development time and (2) to consistently build reusable production software components.

• The Journal of Korea Robotics Society
• /
• v.19 no.1
• /
• pp.106-116
• /
• 2024

Ensuring robust 3D object detection is a core challenge for autonomous driving systems operating in urban environments. To tackle this issue, various 3D representation, including point cloud, voxels, and pillars, have been widely adopted, making use of LiDAR, Camera, and Radar sensors. These representations improved 3D object detection performance, but real-world urban scenarios with unexpected situations can still lead to numerous false positives, posing a challenge for robust 3D models. This paper presents a post-processing algorithm that dynamically adjusts object detection thresholds based on the distance from the ego-vehicle. While conventional perception algorithms typically employ a single threshold in post-processing, 3D models perform well in detecting nearby objects but may exhibit suboptimal performance for distant ones. The proposed algorithm tackles this issue by employing adaptive thresholds based on the distance from the ego-vehicle, minimizing false negatives and reducing false positives in the 3D model. The results show performance enhancements in the 3D model across a range of scenarios, encompassing not only typical urban road conditions but also scenarios involving adverse weather conditions.

• ETRI Journal
• /
• v.45 no.5
• /
• pp.811-821
• /
• 2023

We propose an adaptive unmanned aerial vehicle (UAV)-assisted object recognition algorithm for urban surveillance scenarios. For UAV-assisted surveillance, UAVs are equipped with learning-based object recognition models and can collect surveillance image data. However, owing to the limitations of UAVs regarding power and computational resources, adaptive control must be performed accordingly. Therefore, we introduce a self-adaptive control strategy to maximize the time-averaged recognition performance subject to stability through a formulation based on Lyapunov optimization. Results from performance evaluations on real-world data demonstrate that the proposed algorithm achieves the desired performance improvements.

• Journal of Institute of Control, Robotics and Systems
• /
• v.16 no.8
• /
• pp.766-770
• /
• 2010

In this paper, we present an effective covariance tracking algorithm based on adaptive size changing of tracking window. Recent researches have advocated the use of a covariance matrix of object image features for tracking objects instead of the conventional histogram object models used in popular algorithms. But, according to the general covariance tracking algorithm, it can not deal with the scale changes of the moving objects. The scale of the moving object often changes in various tracking environment and the tracking window(or object kernel) has to be adapted accordingly. In addition, the covariance matrix of moving objects should be adaptively updated considering of the tracking window size. We provide a solution to this problem by segmenting the moving object from the background pixels of the tracking window. Therefore, we can improve the tracking performance of the covariance tracking method. Our several simulations prove the effectiveness of the proposed method.

• International journal of advanced smart convergence
• /
• v.4 no.2
• /
• pp.20-28
• /
• 2015

In this paper, we propose an effective tracking algorithm with an appearance model based on features extracted from a video frame with posture variation and camera view point adaptation by employing the non-adaptive random projections that preserve the structure of the image feature space of objects. The existing online tracking algorithms update models with features from recent video frames and the numerous issues remain to be addressed despite on the improvement in tracking. The data-dependent adaptive appearance models often encounter the drift problems because the online algorithms does not get the required amount of data for online learning. So, we propose an effective tracking algorithm with an appearance model based on features extracted from a video frame.

• Journal of Broadcast Engineering
• /
• v.25 no.3
• /
• pp.362-373
• /
• 2020

Recently, as autonomous vehicles and intelligent CCTV are growing more interest, the efficient object detection is essential technique. The DPM(Deformable Part Models) which is basis of this paper have used a typical object system that represents highly variable objects using mixtures of deformable part for object. Although it shows high detection performance by capturing part shape and configuration of object model, but it is limited to use in real application due to the complicated algorithm. In this paper, instead of image feature pyramid that takes up a large amount of computation in one part of the detector, we propose a method to reduce the computation speed by reconstructing a new image feature pyramid that uses adaptive bilinear interpolation of feature maps obtained on a specific image scale. As a result, the detection performance for object was lowered a little by 2.82%, however, the proposed detection method improved the speed performance by 10% in comparison with original DPM.

• Annual Conference of KIPS
• /
• 2004.05a
• /
• pp.1379-1382
• /
• 2004

The use of a cache for storing and processing of Web objects is becoming larger. Also, many studies on the efficient management of the storing scope of caches are being done. Web caching algorithms have many differences from traditional algorithms. Particularly, heterogeneity of Web objects that are processing units of Web caching, and a variation of Web object reference characteristic with time are the important causes of the decrease the performance of existing algorithms. In this study, we proposed the new web-caching algorithm. A heterogeneity variation of an object can be reduced as the proposed method dividedly managing Web objects and a cache scope with heterogeneity, and it is adaptively reflecting a variation of object reference characteristics with the flowing of time. In the experiments, we verified that the performance of the proposed method was more improved than existing algorithms through the two experiment models which considered heterogeneity of an object.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.9 no.4
• /
• pp.436-443
• /
• 1999

Most of the 3D object reconstruction techniques divide the object into multiplane and approximate the surfaces of the object. The Marching Cubes Algorithm which initializes the mesh structure using a given isovalue. and Delaunay Tetrahedrisation are widely used. Deformable models are well-suited for general object reconstruction because they make little assumptions about the shape to recover and they can reconstruct objects *om various types of datasets. Now, many researchers are studying the reconstruction systems based on a deformable model. In this paper, we propose a novel method for reconstruction of 3D objects. This method, for a 3D object composed of curved planes, compresses the 3D object based on the adaptive simplexmesh technique. It changes the pre-defined mesh structure, so that it may approach to the original object. Also, we redefine the geometric characteristics such as curvatures. As results of simulations, we show reconstruction of the original object with high compression and concentration of vertices towards parts of high curvature in order to optimize the shape description.