• Journal of Advanced Navigation Technology
• /
• v.27 no.2
• /
• pp.190-196
• /
• 2023

Recently, following the development of LIDAR technology which can detect distance from the object, the interest for LIDAR based 3D object detection network is getting higher. Previous networks generate inaccurate localization results due to spatial information loss during voxelization and downsampling. In this study, we propose an attention-based convergence method and a camera-LIDAR convergence system to acquire high-level features and high positional accuracy. First, by introducing the attention method into the Voxel-RCNN structure, which is a grid-based 3D object detection network, the multi-scale sparse 3D convolution feature is effectively fused to improve the performance of 3D object detection. Additionally, we propose the late-fusion mechanism for fusing outcomes in 3D object detection network and 2D object detection network to delete false positive. Comparative experiments with existing algorithms are performed using the KITTI data set, which is widely used in the field of autonomous driving. The proposed method showed performance improvement in both 2D object detection on BEV and 3D object detection. In particular, the precision was improved by about 0.54% for the car moderate class compared to Voxel-RCNN.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.391-401
• /
• 2022

As deep learning-based object detection and recognition research have been developed recently, the scope of application to industry and real life is expanding. But deep learning-based systems in the construction system are still much less studied. Calculating materials in the construction system is still manual, so it is a reality that transactions of wrong volumn calculation are generated due to a lot of time required and difficulty in accurate accumulation. A fast and accurate automatic drawing recognition system is required to solve this problem. Therefore, we propose an AI-based automatic drawing recognition accumulation system that detects and recognizes steel materials in construction drawings. To accurately detect steel materials in construction drawings, we propose data augmentation techniques and spatial attention modules for improving small object detection performance based on YOLOv4. The detected steel material area is recognized by text, and the number of steel materials is integrated based on the predicted characters. Experimental results show that the proposed method increases the accuracy and precision by 1.8% and 16%, respectively, compared with the conventional YOLOv4. As for the proposed method, Precision performance was 0.938. The recall was 1. Average Precision AP_0.5 was 99.4% and AP_0.5:0.95 was 67%. Accuracy for character recognition obtained 99.9.% by configuring and learning a suitable dataset that contains fonts used in construction drawings compared to the 75.6% using the existing dataset. The average time required per image was 0.013 seconds in the detection, 0.65 seconds in character recognition, and 0.16 seconds in the accumulation, resulting in 0.84 seconds.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.18 no.4
• /
• pp.1-11
• /
• 2022

Recently, research on license plate recognition, which is a core technology of an intelligent transportation system(ITS), is being actively conducted. In this paper, we propose a method to extract numbers and characters from low-quality license plate images by applying the YOLOv4 algorithm. YOLOv4 is a one-stage object detection method using convolution neural network including BACKBONE, NECK, and HEAD parts. It is a method of detecting objects in real time rather than the previous two-stage object detection method such as the faster R-CNN. In this paper, we studied a method to directly extract number and character regions from low-quality license plate images without additional edge detection and image segmentation processes. In order to evaluate the performance of the proposed method we experimented with 500 license plate images. In this experiment, 350 images were used for training and the remaining 150 images were used for the testing process. Computer simulations show that the mean average precision of detecting number and character regions on vehicle license plates was about 93.8%.

• Journal of Broadcast Engineering
• /
• v.24 no.5
• /
• pp.802-812
• /
• 2019

With the rapid development in the field of unmanned aerial vehicles(UAVs) and drones, higher request to development of a surveillance system for a drone is putting forward. Since surveillance systems with fixed cameras have a limited range, a development of surveillance systems with a moving camera applicable to PTZ(Pan-Tilt-Zoom) cameras is required. Selecting the features for object plays a critical role in tracking, and the object has to be represented by their shapes or appearances. Considering these conditions, in this paper, an object tracking method with optical flow is introduced to track a tiny drone with a moving camera. In addition, a tracking method combined with kalman filter is proposed to track continuously even when tracking is failed. Experiments are tested on sequences which have a target from the minimal 12 pixels to the maximal 56337 pixels, the proposed method achieves average precision of 175% improvement. Also, experimental results show the proposed method tracks a target which has a size of 12pixels.

• Journal of the Korean Society for Precision Engineering
• /
• v.32 no.7
• /
• pp.633-641
• /
• 2015

In this study, we proposed a three-dimensional (3D) scanning system based on laser-vision technique and rotary mechanism for automatic 3D model reconstruction. The proposed scanning system consists of a laser projector, a camera, and a turntable. For laser-camera calibration a new and simple method was proposed. 3D point cloud data of the surface of scanned object was fully collected by integrating extracted laser profiles, which were extracted from laser stripe images, corresponding to rotary angles of the rotary mechanism. The obscured laser profile problem was also solved by adding an addition camera at another viewpoint. From collected 3D point cloud data, the 3D model of the scanned object was reconstructed based on facet-representation. The reconstructed 3D models showed effectiveness and the applicability of the proposed 3D scanning system to 3D model-based applications.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.6
• /
• pp.907-917
• /
• 2023

Recently, a self-driving car have applied deep learning technology to advanced driver assistance system can provide convenience to drivers, but it is shown deep that learning technology is vulnerable to adversarial evasion attacks. In this paper, we performed five adversarial evasion attacks, including MI-FGSM(Momentum Iterative-Fast Gradient Sign Method), targeting the object detection algorithm YOLOv5 (You Only Look Once), and measured the object detection performance in terms of mAP(mean Average Precision). In particular, we present a method applying morphology operations for YOLO to detect objects normally by removing noise and extracting boundary. As a result of analyzing its performance through experiments, when an adversarial attack was performed, YOLO's mAP dropped by at least 7.9%. The YOLO applied our proposed method can detect objects up to 87.3% of mAP performance.

• Proceedings of the KSRS Conference
• /
• 2004.10a
• /
• pp.125-127
• /
• 2004

High-resolution satellite images offer abundance information of the earth surface for remote sensing applications. The information includes geometry, texture and attribute characteristic. The pixel-based image classification can't satisfy high-resolution satellite image's classification precision and produce large data redundancy. Object-oriented information extraction not only depends on spectrum character, but also use geometry and structure information. It can provide an accessible and truly revolutionary approach. Using Beijing Spot 5 high-resolution image and object-oriented classification with the eCognition software, we accomplish the cultures' precise classification. The test areas have five culture types including water, vegetation, road, building and bare lands. We use nearest neighbor classification and appraise the overall classification accuracy. The average of five species reaches 0.90. All of maximum is 1. The standard deviation is less than 0.11. The overall accuracy can reach $95.47\%.$ This method offers a new technology for high-resolution satellite images' available applications in remote sensing culture classification.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.18 no.11
• /
• pp.705-710
• /
• 2017

This paper proposes an estimation method of populations of Grapholita molestas using object segmentation and an SVM classifier in the moth images. Object segmentation and moth classification were performed on images of Grapholita molestas moth acquired on a pheromone trap equipped in an orchard. Object segmentation consisted of pre-processing, thresholding, morphological filtering, and object labeling process. The classification of Grapholita molestas in the moth images consisted of the training and classification of an SVM classifier and estimation of the moth populations. The object segmentation simplifies the moth classification process by segmenting the individual objects before passing an input image to the SVM classifier. The image blocks were extracted around the center point and principle axis of the segmented objects, and fed into the SVM classifier. In the experiments, the proposed method performed an estimation of the moth populations for 10 moth images and achieved an average estimation precision rate of 97%. Therefore, it showed an effective monitoring method of populations of Grapholita molestas in the orchard. In addition, the mean processing time of the proposed method and sliding window technique were 2.4 seconds and 5.7 seconds, respectively. Therefore, the proposed method has a 2.4 times faster processing time than the latter technique.

• Journal of the HCI Society of Korea
• /
• v.1 no.1
• /
• pp.45-52
• /
• 2006

This paper presents an image-space algorithm to real-time collision detection, which is run completely by GPU. For a single object or for multiple objects with no collision, the front and back faces appear alternately along the view direction. However, such alternation is violated when objects collide. Based on these observations, the algorithm propose the depth peeling method which renders the minimal surface of objects, not whole surface, to find colliding. The Depth peeling method utilizes the state-of-the-art functionalities of GPU such as framebuffer object, vertexbuffer object, and occlusion query. Combining these functions, multi-pass rendering and context switch can be done with low overhead. Therefore proposed approach has less rendering times and rendering overhead than previous image-space collision detection. The algorithm can handle deformable objects and complex objects, and its precision is governed by the resolution of the render-target-texture. The experimental results show the feasibility of GPU-based collision detection and its performance gain in real-time applications such as 3D games.

• International Journal of Precision Engineering and Manufacturing
• /
• v.4 no.2
• /
• pp.64-70
• /
• 2003

A new method using a transfer function is introduced in the present paper for analyzing the motion errors of hydrostatic bearing tables. The relationship between film reaction force in a single-side hydrostatic pad and the form error of guide rail is derived at various spatial frequencies by finite element analysis, and it is expressed as a transfer function. This transfer function clarifies so called 'the averaging effect of an oil film' quantitively. It is found that the amplitude of film force is reduced as the spatial frequency increases or the relative width of the pocket is reduced. The motion errors of a multi pad type table are estimated using a transfer function, the form errors of a guide rail and the geometric relationship between the pads. The method is named as the Transfer Function Method (TFM). The motion errors calculated by the TFM show good agreement with the motion errors calculated by the Multi Pad Method considering the entire table as an analysis object. From the results, it is confirmed that the proposed TFM is very effective to analyze the motion errors of hydrostatic tables.