• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.10a
• /
• pp.422-424
• /
• 2021

In this paper, it leads the approach of fusing multiple RGB cameras for visual objects recognition based on deep learning with convolution neural network and 3D Light Detection and Ranging (LiDAR) to observe the environment and match into a 3D world in estimating the distance and position in a form of point cloud map. The goal of perception in multiple cameras are to extract the crucial static and dynamic objects around the autonomous vehicle, especially the blind spot which assists the AV to navigate according to the goal. Numerous cameras with object detection might tend slow-going the computer process in real-time. The computer vision convolution neural network algorithm to use for eradicating this problem use must suitable also to the capacity of the hardware. The localization of classified detected objects comes from the bases of a 3D point cloud environment. But first, the LiDAR point cloud data undergo parsing, and the used algorithm is based on the 3D Euclidean clustering method which gives an accurate on localizing the objects. We evaluated the method using our dataset that comes from VLP-16 and multiple cameras and the results show the completion of the method and multi-sensor fusion strategy.

• 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.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.10
• /
• pp.19-28
• /
• 2022

In this paper, we propose the Detector Scheduler which determines the best tracking-by-detection (TBD) mechanism to perform real-time high-accurate multi-object tracking (MOT). The Detector Scheduler determines whether to run a detector by measuring the dissimilarity of features between different frames. Furthermore, we propose a self-supervision method to learn the Detector Scheduler with tracking results since it is difficult to generate ground truth (GT) for learning the Detector Scheduler. Our proposed self-supervision method generates pseudo labels on whether to run a detector when the dissimilarity of the object cardinality or appearance between frames increases. To this end, we propose the Detector Scheduling Loss to learn the Detector Scheduler. As a result, our proposed method achieves real-time high-accurate multi-object tracking by boosting the overall tracking speed while keeping the tracking accuracy at most.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.13 no.6
• /
• pp.1263-1268
• /
• 2018

In this paper, a video analysis is proposed to implement video surveillance system with deep learning object detection and probabilistic data association filter for tracking multiple objects, and suggests its implementation using GPU. The proposed video analysis technique involves object detection and object tracking sequentially. The deep learning network architecture uses ResNet for object detection and applies probabilistic data association filter for multiple objects tracking. The proposed video analysis technique can be used to detect intruders illegally trespassing any restricted area or to count the number of people entering a specified area. As a results of simulations and experiments, 48 channels of videos can be analyzed at a speed of about 27 fps and real-time video analysis is possible through RTSP protocol.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2015.01a
• /
• pp.137-138
• /
• 2015

본 논문에서는 연속된 영상에서 다중해상도 에지정보의 차이를 이용하여 이동하는 물체를 탐지하고 계수하는 시스템을 제안한다. 연속적으로 입력되는 영상에 대하여 이산 웨이블릿 연산을 수행하여 다중해상도 에지를 추출하고, 인접한 프레임 사이의 다중해상도 에지 차이를 이용하여 이동물체를 추출한다. 가중치가 부여된 유클리디언 거리를 이용하여 물체를 추적한 다음, 칼만필터를 이용하여 물체 궤적의 위치 정보를 보정한다. 마지막으로, 관심영역에 대한 물체 궤적의 상대적인 위치를 고려하여 이동물체를 계수한다.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.13 no.5
• /
• pp.35-49
• /
• 2014

This paper proposes the on-road object detection and classification algorithm by using a detection system consisting of only laser scanners. Each sensor data acquired by the laser scanner is fused with a grid map and the measurement error and spot spaces are corrected using a labeling method and dilation operation. Fuzzy method which uses the object information (length, width) as input parameters can classify the objects such as a pedestrian, bicycle and vehicle. In this way, the accuracy of the detection system is increased. Through experiments for some scenarios in the real road environment, the performance of the proposed detection and classification system for the actual objects is demonstrated through the comparison with the actual information acquired by GPS-RTK.

• Ocean Systems Engineering
• /
• v.10 no.4
• /
• pp.435-449
• /
• 2020

This paper proposes a method to estimate the underwater target object's yaw angle using a sonar image. A simulator modeling imaging mechanism of a sonar sensor and a generative adversarial network for style transfer generates realistic template images of the target object by predicting shapes according to the viewing angles. Then, the target object's yaw angle can be estimated by comparing the template images and a shape taken in real sonar images. We verified the proposed method by conducting water tank experiments. The proposed method was also applied to AUV in field experiments. The proposed method, which provides bearing information between underwater objects and the sonar sensor, can be applied to algorithms such as underwater localization or multi-view-based underwater object recognition.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.12
• /
• pp.4439-4455
• /
• 2021

Compared with two-stage object detection algorithms, one-stage algorithms provide a better trade-off between real-time performance and accuracy. However, these methods treat the intermediate features equally, which lacks the flexibility to emphasize meaningful information for classification and location. Besides, they ignore the interaction of contextual information from different scales, which is important for medium and small objects detection. To tackle these problems, we propose an image pyramid network based on dual attention mechanism (DAIPNet), which builds an image pyramid to enrich the spatial information while emphasizing multi-scale informative features based on dual attention mechanisms for one-stage object detection. Our framework utilizes a pre-trained backbone as standard detection network, where the designed image pyramid network (IPN) is used as auxiliary network to provide complementary information. Here, the dual attention mechanism is composed of the adaptive feature fusion module (AFFM) and the progressive attention fusion module (PAFM). AFFM is designed to automatically pay attention to the feature maps with different importance from the backbone and auxiliary network, while PAFM is utilized to adaptively learn the channel attentive information in the context transfer process. Furthermore, in the IPN, we build an image pyramid to extract scale-wise features from downsampled images of different scales, where the features are further fused at different states to enrich scale-wise information and learn more comprehensive feature representations. Experimental results are shown on MS COCO dataset. Our proposed detector with a 300 × 300 input achieves superior performance of 32.6% mAP on the MS COCO test-dev compared with state-of-the-art methods.

• The Journal of Korean Institute of Information Technology
• /
• v.17 no.5
• /
• pp.13-20
• /
• 2019

Recently, object detection technology based on CNN has been actively studied. Object detection technology is used as an important technology in autonomous vehicles, intelligent image analysis, and so on. In this paper, we propose a rotation change robust object detection system by applying MI-FL (Moment Invariant-Feature Layer) to SSD (Single Shot Multibox Detector) which is one of CNN-based object detectors. First, the features of the input image are extracted based on the VGG network. Then, a total of six feature layers are applied to generate bounding boxes by predicting the location and type of object. We then use the NMS algorithm to get the bounding box that is the most likely object. Once an object bounding box has been determined, the invariant moment feature of the corresponding region is extracted using MI-FL, and stored and learned in advance. In the detection process, it is possible to detect the rotated image more robust than the conventional method by using the previously stored moment invariant feature information. The performance improvement of about 4 ~ 5% was confirmed by comparing SSD with existing SSD and MI-FL.

• Journal of Biosystems Engineering
• /
• v.23 no.4
• /
• pp.381-390
• /
• 1998

An algorithm to extract the 3-D geometric information of a static object was developed using a set of 2-D computer vision system and a laser structured lighting device. As a structured light pattern, multi-parallel lines were used in the study. The proposed algorithm was composed of three stages. The camera calibration, which determined a coordinate transformation between the image plane and the real 3-D world, was performed using known 6 pairs of points at the first stage. Then, utilizing the shifting phenomena of the projected laser beam on an object, the height of the object was computed at the second stage. Finally, using the height information of the 2-D image point, the corresponding 3-D information was computed using results of the camera calibration. For arbitrary geometric objects, the maximum error of the extracted 3-D feature using the proposed algorithm was less than 1~2mm. The results showed that the proposed algorithm was accurate for 3-D geometric feature detection of an object.