• International Journal of Computer Science & Network Security
• /
• v.24 no.5
• /
• pp.129-134
• /
• 2024

Object Detection is an emerging technology in the field of Computer Vision and Image Processing that deals with detecting objects of a particular class in digital images. It has considered being one of the complicated and challenging tasks in computer vision. Earlier several machine learning-based approaches like SIFT (Scale-invariant feature transform) and HOG (Histogram of oriented gradients) are widely used to classify objects in an image. These approaches use the Support vector machine for classification. The biggest challenges with these approaches are that they are computationally intensive for use in real-time applications, and these methods do not work well with massive datasets. To overcome these challenges, we implemented a Deep Learning based approach Convolutional Neural Network (CNN) in this paper. The Proposed approach provides accurate results in detecting objects in an image by the area of object highlighted in a Bounding Box along with its accuracy.

• KIPS Transactions on Software and Data Engineering
• /
• v.3 no.9
• /
• pp.361-368
• /
• 2014

The LPR(License plate recognition) system has been developed to efficient control for complex traffic environment and currently be used in many places. However, because of light, noise, background changes, environmental changes, damaged plate, it only works limited environment, so it is difficult to use in real-time. This paper presents a heuristic segmentation algorithm for robust to noise and illumination changes and introduce a real-time license plate recognition system using it. In first step, We detect the plate utilized Haar-like feature and Adaboost. This method is possible to rapid detection used integral image and cascade structure. Second step, we determine the type of license plate with adaptive histogram equalization, bilateral filtering for denoise and segment accurate character based on adaptive threshold, pixel projection and associated with the prior knowledge. The last step is character recognition that used histogram of oriented gradients (HOG) and multi-layer perceptron(MLP) for number recognition and support vector machine(SVM) for number and Korean character classifier respectively. The experimental results show license plate detection rate of 94.29%, license plate false alarm rate of 2.94%. In character segmentation method, character hit rate is 97.23% and character false alarm rate is 1.37%. And in character recognition, the average character recognition rate is 98.38%. Total average running time in our proposed method is 140ms. It is possible to be real-time system with efficiency and robustness.

• Journal of Internet Computing and Services
• /
• v.20 no.4
• /
• pp.91-102
• /
• 2019

In autonomous driving systems, the ability to classify pedestrians in images captured by cameras is very important for pedestrian safety. In the past, after extracting features of pedestrians with HOG(Histogram of Oriented Gradients) or SIFT(Scale-Invariant Feature Transform), people classified them using SVM(Support Vector Machine). However, extracting pedestrian characteristics in such a handcrafted manner has many limitations. Therefore, this paper proposes a method to classify pedestrians reliably and effectively using CNN's(Convolutional Neural Network) deep features and transfer learning. We have experimented with both the fixed feature extractor and the fine-tuning methods, which are two representative transfer learning techniques. Particularly, in the fine-tuning method, we have added a new scheme, called M-Fine(Modified Fine-tuning), which divideslayers into transferred parts and non-transferred parts in three different sizes, and adjusts weights only for layers belonging to non-transferred parts. Experiments on INRIA Person data set with five CNN models(VGGNet, DenseNet, Inception V3, Xception, and MobileNet) showed that CNN's deep features perform better than handcrafted features such as HOG and SIFT, and that the accuracy of Xception (threshold = 0.5) isthe highest at 99.61%. MobileNet, which achieved similar performance to Xception and learned 80% fewer parameters, was the best in terms of efficiency. Among the three transfer learning schemes tested above, the performance of the fine-tuning method was the best. The performance of the M-Fine method was comparable to or slightly lower than that of the fine-tuningmethod, but higher than that of the fixed feature extractor method.

• Journal of the Korea Society of Computer and Information
• /
• v.22 no.12
• /
• pp.79-85
• /
• 2017

In this paper, we propose a CNN-based swimmer detection algorithm. Every year, water safety accidents have been occurred frequently, and accordingly, intelligent video surveillance systems are being developed to prevent accidents. Intelligent video surveillance system is a real-time system that detects objects which users want to do. It classifies or detects objects in real-time using algorithms such as GMM (Gaussian Mixture Model), HOG (Histogram of Oriented Gradients), and SVM (Support Vector Machine). However, HOG has a problem that it cannot accurately detect the swimmer in a complex and dynamic environment such as a beach. In other words, there are many false positives that detect swimmers as waves and false negatives that detect waves as swimmers. To solve this problem, in this paper, we propose a swimmer detection algorithm using CNN (Convolutional Neural Network), specialized for small object sizes, in order to detect dynamic objects and swimmers more accurately and efficiently in complex environment. The proposed CNN sets the size of the input image and the size of the filter used in the convolution operation according to the size of objects. In addition, the aspect ratio of the input is adjusted according to the ratio of detected objects. As a result, experimental results show that the proposed CNN-based swimmer detection method performs better than conventional techniques.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.12
• /
• pp.6000-6017
• /
• 2018

This paper presents a new descriptor, named Histograms of Prominent Edge Directions (HPED), for the recognition of facial expressions in a person-independent environment. In this paper, we raise the issue of sampling error in generating the code-histogram from spatial regions of the face image, as observed in the existing descriptors. HPED describes facial appearance changes based on the statistical distribution of the top two prominent edge directions (i.e., primary and secondary direction) captured over small spatial regions of the face. Compared to existing descriptors, HPED uses a smaller number of code-bins to describe the spatial regions, which helps avoid sampling error despite having fewer samples while preserving the valuable spatial information. In contrast to the existing Histogram of Oriented Gradients (HOG) that uses the histogram of the primary edge direction (i.e., gradient orientation) only, we additionally consider the histogram of the secondary edge direction, which provides more meaningful shape information related to the local texture. Experiments on popular facial expression datasets demonstrate the superior performance of the proposed HPED against existing descriptors in a person-independent environment.

• Journal of Institute of Control, Robotics and Systems
• /
• v.22 no.2
• /
• pp.110-116
• /
• 2016

Autonomous vehicles must obey the traffic laws in order to drive actual roads. Traffic signs erected at the side of roads explain the road traffic information or regulations. Therefore, traffic sign recognition is necessary for the autonomous vehicles. In this paper, color characteristics are first considered to detect traffic sign candidates. Subsequently, we establish HOG (Histogram of Oriented Gradients) features from the detected candidate and recognize the traffic sign through a SVM (Support Vector Machine). However, owing to various circumstances, such as changes in weather and lighting, it is difficult to recognize the traffic signs robustly using only SVM. In order to solve this problem, we propose a tracking algorithm with RANSAC-based motion estimation. Using two-point motion estimation, inlier feature points within the traffic sign are selected and then the optimal motion is calculated with the inliers through a bundle adjustment. This approach greatly enhances the traffic sign recognition performance.

• ETRI Journal
• /
• v.35 no.6
• /
• pp.1021-1028
• /
• 2013

Automatic facial expression recognition is a widely studied problem in computer vision and human-robot interaction. There has been a range of studies for representing facial descriptors for facial expression recognition. Some prominent descriptors were presented in the first facial expression recognition and analysis challenge (FERA2011). In that competition, the Local Gabor Binary Pattern Histogram Sequence descriptor showed the most powerful description capability. In this paper, we introduce hybrid facial representations for facial expression recognition, which have more powerful description capability with lower dimensionality. Our descriptors consist of a block-based descriptor and a pixel-based descriptor. The block-based descriptor represents the micro-orientation and micro-geometric structure information. The pixel-based descriptor represents texture information. We validate our descriptors on two public databases, and the results show that our descriptors perform well with a relatively low dimensionality.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.9
• /
• pp.3712-3729
• /
• 2020

Smoke detection is helpful for early fire detection. With its large coverage area and low cost, vision-based smoke detection technology is the main research direction of outdoor smoke detection. We propose a two-stage smoke detection method combined with block Deep Normalization and Convolutional Neural Network (DNCNN) and visual change image. In the first stage, each suspected smoke region is detected from each frame of the images by using block DNCNN. According to the physical characteristics of smoke diffusion, a concept of visual change image is put forward in this paper, which is constructed by the video motion change state of the suspected smoke regions, and can describe the physical diffusion characteristics of smoke in the time and space domains. In the second stage, the Support Vector Machine (SVM) classifier is used to classify the Histogram of Oriented Gradients (HOG) features of visual change images of the suspected smoke regions, in this way to reduce the false alarm caused by the smoke-like objects such as cloud and fog. Simulation experiments are carried out on two public datasets of smoke. Results show that the accuracy and recall rate of smoke detection are high, and the false alarm rate is much lower than that of other comparison methods.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.38 no.4
• /
• pp.345-354
• /
• 2014

In this paper, we propose a real-time algorithm for estimating the relative position and velocity of a person with respect to a robot using a Kinect sensor and an extended Kalman filter (EKF). Additionally, we propose an algorithm for controlling the robot in the proximity of a person in a variety of modes. The algorithm detects the head and shoulder regions of the person using a histogram of oriented gradients (HOG) and a support vector machine (SVM). The EKF algorithm estimates the relative positions and velocities of the person with respect to the robot using data acquired by a Kinect sensor. We tested the various modes of proximity movement for a human in indoor situations. The accuracy of the algorithm was verified using a motion capture system.

• The Journal of Korean Institute of Information Technology
• /
• v.16 no.11
• /
• pp.85-95
• /
• 2018

With the popularity of smartphones, most peoples have been using mobile cameras to capture photographs. However, due to insufficient amount of lights in a low lighting condition, unwanted noises can be generated during image acquisition. To remove the noise, a method of using deep convolutional neural networks is introduced. However, this method still lacks the ability to describe textures and edges, even though it has made significant progress in terms of visual quality performance. Therefore, in this paper, the HOG (Histogram of Oriented Gradients) images that contain information about edge orientations are used. More specifically, a method of learning deep convolutional neural networks is proposed by stacking noise and HOG images into an input tensor. Experiment results confirm that the proposed method not only can obtain excellent result in visual quality evaluations, compared to conventional methods, but also enable textures and edges to be improved visually.