• Smart Structures and Systems
• /
• v.24 no.6
• /
• pp.745-757
• /
• 2019

Road surface deteriorations such as potholes have caused motorists heavy monetary damages every year. However, effective road condition monitoring has been a continuing challenge to road owners. Depth cameras have a small field of view and can be easily affected by vehicle bouncing. Traditional image processing methods based on algorithms such as segmentation cannot adapt to varying environmental and camera scenarios. In recent years, novel object detection methods based on deep learning algorithms have produced good results in detecting typical objects, such as faces, vehicles, structures and more, even in scenarios with changing object distances, camera angles, lighting conditions, etc. Therefore, in this study, a Deep Learning Pothole Detector (DLPD) based on the deep region-based convolutional neural network is proposed for autonomous detection of potholes from images. About 900 images with potholes and road surface conditions are collected and divided into training and testing data. Parameters of the network in the DLPD are calibrated based on sensitivity tests. Then, the calibrated DLPD is trained by the training data and applied to the 215 testing images to evaluate its performance. It is demonstrated that potholes can be automatically detected with high average precision over 93%. Potholes can be differentiated from manholes by training and applying a manhole-pothole classifier which is constructed using the convolutional neural network layers in DLPD. Repeated detection of the same potholes can be prevented through feature matching of the newly detected pothole with previously detected potholes within a small region.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.41 no.8
• /
• pp.843-850
• /
• 2016

This paper introduces a novel modulation scheme for vehicular communication in taking advantage of existing LED lights available on a car. Our proposed 2-Phase Shift Keying (2-PSK) is a spatial modulation approach in which a pair of LED light sources in a car (either rear LEDs or front LEDs) is used as a transmitter. A typical camera (i.e. low frame rate at no greater than 30fps) that either a global shutter camera or a rolling shutter camera can be used as a receiver. The modulation scheme is a part of our Image Sensor Communication proposal submitted to IEEE 802.15.7r1 (TG7r1) recently. Also, a neural network approach is applied to improve the performance of LEDs detection and decoding under the noisy situation. Later, some analysis and experiment results are presented to indicate the performance of our system

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.3 no.4
• /
• pp.867-873
• /
• 1999

Camera calibration is a procedure which calculates internal and external parameters of a camera with the Down world coordinates of the control points. Accurate camera calibration is required for achieving accurate visual measurements. In this paper, we propose a simple and flexible camera calibration using neural networks which doesn't require a special knowledge of 3D geometry and camera optics. There are some applications which are not in need of the values of the internal and external parameters. The proposed method is very useful to these applications. Also, the proposed camera calibration has advantage that resolves the ill-condition as object plane is near parallel image plane. The ill-condition is frequently met in product inspection. For little more accurate calibration, acquired image is divided into two regions according to radial distortion of lens and neural network is applied to each region. Experimental results and comparison with Tsai's algorithm prove the validity of the proposed camera calibration.

• Journal of IKEEE
• /
• v.21 no.3
• /
• pp.268-271
• /
• 2017

In this paper, we propose the development of high resolution $360^{\circ}$ omnidirectional IP camera with 12 million pixels. The proposed 12-megapixel high-resolution $360^{\circ}$ omnidirectional IP camera consists of a lens unit with $360^{\circ}$ omnidirectional viewing angle and a 12-megapixel high-resolution IP camera unit. The lens section of $360^{\circ}$ omnidirectional viewing angle adopts the isochronous lens design method and the catadioptric facet production method to obtain the image without peripheral distortion which is inevitably generated in the fisheye lens. The 12 megapixel high-resolution IP camera unit consists of a CMOS sensor & ISP unit, a DSP unit, and an I / O unit, and converts the image input to the camera into a digital image to perform image distortion correction, image correction and image compression And then transmits it to the NVR (Network Video Recorder). In order to evaluate the performance of the proposed 12-megapixel high-resolution $360^{\circ}$ omnidirectional IP camera, 12.3 million pixel image efficiency, $360^{\circ}$ omnidirectional lens angle of view, and electromagnetic certification standard were measured.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2001.12a
• /
• pp.263-266
• /
• 2001

It has been integrated into several navigation systems. This paper shows that system recognizes difficult indoor roads and open area without any specific mark such as painted guide line or tape. In this method, Robot navigates with visual sensors, which uses visual information to navigate itself along the road. An Artificial Neural Network System was used to decide where to move. It is designed with USB web camera as visual sensor.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 1995.10a
• /
• pp.341-343
• /
• 1995

In this paper, an algorithm is presented to recogniz the driving direction of a vehicle and obstacles in front of it based on highway road image. The algorithm employs a neural network with 27 sub sets obtained from the road image as its input. The outputs include the direction of the vehicle movement and presence or absence of obstacles. The road image, obtained by a video camera, was digitized and processed by a personal computer equipped with an image processing board.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.2033-2036
• /
• 2004

A novel approach to estimate the real time moving trajectory of an object is proposed in this paper. The object position is obtained from the image data of a CCD camera, while a state estimator predicts the linear and angular velocities of the moving object. To overcome the uncertainties and noises residing in the input data, a Kalman filter and neural networks are utilized. Since the Kalman filter needs to approximate a non-linear system into a linear model to estimate the states, there always exist errors as well as uncertainties again. To resolve this problem, the neural networks are adopted in this approach, which have high adaptability with the memory of the input-output relationship. Kohonen Network(Self-Organized Map) is selected to learn the motion trajectory since it is spatially oriented. The superiority of the proposed algorithm is demonstrated through the real experiments.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.2 no.1
• /
• pp.1-8
• /
• 2002

It has been integrated into several navigation systems. This paper shows that system recognizes difficult indoor roads without any specific marks such as painted guide line or tape. In this method the robot navigates with visual sensors, which uses visual information to navigate itself along the read. The Neural Network System was used to learn driving pattern and decide where to move. In this paper, I will present a vision-based process for AMR(Autonomous Mobile Robot) that is able to navigate on the indoor read with simple computation. We used a single USB-type web camera to construct smaller and cheaper navigation system instead of expensive CCD camera.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.7
• /
• pp.3100-3116
• /
• 2016

With the rapid development of urban traffic system and fast increasing of vehicle numbers, the traditional centralized ways to generate the source-destination shortest path in terms of travel time(the optimal path) encounter several problems, such as high server pressure, low query efficiency, roads state without in-time updating. With the widespread use of smart cameras in the urban traffic and surveillance system, this paper maps the optimal path finding problem in the dynamic road network to the shortest routing problem in the smart camera networks. The proposed distributed optimal path generation algorithm employs the delay routing and caching mechanism. Real-time route update is also presented to adapt to the dynamic road network. The test result shows that this algorithm has advantages in both query time and query packet numbers.

• Proceedings of the KIEE Conference
• /
• 2002.07d
• /
• pp.2780-2782
• /
• 2002

In this paper Valentine Braitenberg structure based sensory motor model for object tracking control system was proposed. Conventional model based control schemes are require highly non-linear mathematical models, which require long computational time to solve complex high order equations. Contrast to conventional models proposed system simply link signal data from camera directly to the inputs of neural network, and outputs of network are directly fed into input of motor driver of camera. With simple structure of sensory motor model, real time tracking control system for dynamic object was realized successfully, and the implementation of sensory motor model can overcome the limitation of model-based control schemes.