JeongMok Ha

Cost Aggregation Table: Cost Aggregation Method Using Summed Area Table Scheme for Dense Stereo Correspondence

The stereo matching algorithms usually do not satisfy the performance of both accuracy and complexity. The semi-global matching (SGM) is the most efficient method among the stereo matching algorithms. However, it still does not utilize the whole image information to estimate the disparity values. In this paper, we proposed a new concept of the cost aggregation method using summed area table (SAT) scheme. The SAT is the efficient algorithm for summing intensities on the rectangular area in the image. Using four kinds of the SAT arrays (we called these arrays cost aggregation table (CAT)), it is possible to estimate the disparity values with aggregating every cost in the image. We tested our algorithm using the KITTI vision benchmark suite, the result shows that our algorithm is superior for disparity accuracy compared to the SGM. We expect that the CAT can be an alternative cost aggregation method to the SGM in the near future.

Image synthesis and occlusion removal of intermediate views by stereo matching

We suggest an efficient real-time method for removing occlusions when synthesizing intermediate views from two stereo images by stereo matching algorithm. The proposed method can carry out the correction of noise resulting from stereo matching, removal of occlusions to achieve high fidelity rendering, and parallel computing to achieve real-time synthesis. The proposed algorithm is compared with state-of-the-art algorithms to show its improved performance in terms of rendering and computational speed.

LED Traffic Sign Detection Using Rectangular Hough Transform

In this paper, a new Advanced Driver Assistant System (ADAS) system for LED traffic sign detection algorithm using rectangle shape based on a windowed hough transform and feature based optimization was presented. We used two character to detect the LED traffic sign. One is LED traffic sign have rectangle shape, another is intensity feature of LED traffic sign. After extracting the candidates of LED traffic signs, our proposed system classify the positive and negative rectangle candidate as a LED traffic sign. Under this flow, we can finally obtained LED traffic sign from real road scene that include LED traffic sign. Our proposed technique was tested in 87 number of real road scene that include LED traffic signs. We can find the 368 number of LED traffic signs of existing 430 number of LED traffic signs. The detection ratio is 85.37%. Algorithm proposed in this paper is very meaningful as a first attempt to detect the LED traffic signs.Detection ratio also reasonable to recognize the traffic sign in the next step of Traffic Sign Recognition (TSR).

Polygonal Symmetry Transform for Detecting Rectangular Traffic Signs

We present a new symmetry transform finding rectangles based on a gradient and corner information of images. We succeed the Fast Radial Symmetry Transform (FRST) finding radial objects with low cost and high accuracy. But FRST focused on using image gradient to find small circle shapes so that FRST can not detect larger polygonal shapes well. Our transform used image gradient and image corner to construct polygonal symmetry maps and extract regions of interest from images. We referred the set of polygon features as the Polygonal Symmetry Transform (PST). Then PST has large coverage with the advantages of FRST. To verify PST, we designed a polygon detectors which finds the road traffic sign such as radial and polygonal shapes. Our detector extracted regions of interest from real road images and calculated performances including detection rate for traffic signs and processing time per image in our experiments.

LED traffic sign detection with the fast radial symmetric transform and symmetric shape detection

We present a method that can detect a circular traffic signs and Light Emitting Diode (LED) traffic signs by combining fast radial symmetric transform and a proposed symmetric shape detection method. The proposed method can choose the appropriate size and sign type by computing the confidence of given image to each pre-defined signs, In a real road video that includes LED signs and circular traffic signs, our algorithm scored 94.3 % detection ratio and 0.33 false positive per frame; these results are acceptable in advanced driver assistance systems.

An improved 2D cost aggregation method for advanced driver assistance systems

In advanced driver assistance systems, the stereo matching algorithm is the key resource to obtain depth information of outdoor scenes. Semi-Global Matching (SGM) is currently the most efficient stereo matching algorithm for outdoor environments. However, because the number of pixels is large, SGM uses only a subset of them when estimating the disparity of a pixel. To overcome this limitation, Cost Aggregation Table (CAT) was proposed which uses two-dimensional cost aggregation so as to utilize whole image information. In this paper, we propose improved global 2D cost aggregation methods by loosening aggregation constraints. It aggregates every cost in the whole image to estimate each disparity. Although our method aggregates every cost in the image, the computational complexity is the same as that of SGM and CAT. The proposed cost aggregation method achieves superior disparity accuracy compared to the SGM.

Moving Object Detection Using Energy Model and Particle Filter for Dynamic Scene

We proposed an algorithm that uses an energy model with smoothness assumption to identify a moving object by using optical flow, and uses a particle filter with a proposed observation and dynamic model to track the object. The algorithm is based on the assumption that the dominant motion is background flow and that foreground flow is separated from the background flow. The energy model provides the initial label foreground object well, and minimizes the number of noise pixels that are included in the bounding box. The tracking part uses HOG-3 as an observation model, and optical flow as the dynamic model. This combination of models improves the accuracy of tracking results. In experiments on challenging data set that have no initial labels, the algorithm achieved meaningful accuracy compared to a state-of-the-art technique that needs initial labels.

Single Image Haze Removal Using Single Pixel Approach Based on Dark Channel Prior with Fast Filtering

We propose a fast image-dehazing method that uses a pixelwise dark channel prior (PDCP). We argue that the neighbor minimum filter in the dark channel prior (DCP) is not crucial in image dehazing. We prove this assertion in experiments that compare dehazed images and histograms of dark channel images obtained using PDCP and DCP. To refine the transmission map in the Koschmieder model of the degradation of a hazy image, we use a fast guided filter to replace the soft matting (SM) used in DCP, because SM is the main reason for the slowness of DCP. The proposed algorithm is faster than existing methods, but achieves similar dehazing. This new method is useful for applications that require fast dehazing.

Moving Object Detection Using SIFT Matching on Three Frames for Advanced Driver Assistance Systems

Detecting moving objects in a dynamic scene is a difficult task in computer vision. We propose a moving object detection algorithm for advanced driver assistance systems that uses only images from a monocular camera. To distinguish moving objects from standing objects when the camera is moving, we used an epipolar line constraint and an optical flow constraint. When evaluated using the KITTI scene flow 2015 dataset, the proposed algorithm detected moving objects in the image successfully even when the monocular camera was moving. The runtime of the proposed algorithm is < 1 s, so it is feasible for practical uses.

A fast scanning based message receiving method on four directed acyclic subgraphs

A fast MAP inference method for the general discrete labeling problems.A new graphical model that consists of four directed acyclic subgraphs (DAS).Message receiving inference based on four DAS structure.Four scanning method that utilizes message receiving inference.Extremely high speed and competitive accuracy compared to other MAP inference. We propose a message-receiving algorithm on a Directed Acyclic Subgraph (DAS) structure to approximate the solution of general labeling problems extremely quickly. The algorithm divides a graph into four subgraphs to get a joint distribution of all nodes, then passes messages in two fixed directions as inference on DASs. Message receiving is a modified version of message passing. When receiving messages on DAS structure, labeling results can be obtained after just four scans. The proposed algorithm was evaluated by using it to perform three labeling decision applications (binary segmentation, image denoising, and stereo matching). Compared to other highly-accurate iterative algorithms ( α -expansion, α - β swap, tree-reweighted message passing, sum-product belief propagation, max-product belief propagation, and FastPD), the proposed algorithm shows competitive accuracy but requires much less computational time. The proposed algorithm is appropriate for applications in which iterative schemes are undesirable, but which must get reliable labeling results within a limited time.