Home | Login
Lectures       Previous announcements
Select year: 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017
Seminars in 2011
    This paper presents a monocular machine vision system capable of detecting vehicles in front or behind of our own vehicle. The system consists of two main steps: 1) generation of candidates with respect to a vehicle by analyzing textures, 2) verification of the candidates by an appearance-based method using the AdaBoost learning algorithm. The vehicle candidates are generated by exploiting the facts that a vehicle has vertical and horizontal lines, and furthermore the rear and frontal shapes of a vehicle show symmetry. The proposed system is proven to be effective through experiments under various traffic scenarios.
    Abstract-This paper describes some development work of a system for elder care configured around sensors and robotic environment. A few sub-systems have been amalgamate3d to achieve the desired objective. The integrated system is able to support people who wish to live alone but, because of old age, ill health or disability, are at risk and are a cause of concern for their family and friends. The first part of the integrated system, named Selective Activity Monitoring(SAM) system, works on the principle of using sensor units (SU) to monitor various electrical appliances in a house. Rules, based on the daily activities of a person, are defined for the appliances to be monitored. The system can detect violation of the rules and generate an alarm. The rules are flexible and can be user-defined. The second part of the system is a low-cost Physiological Parameters Monitoring (PPM) system, called Medicmate, which can be used to monitor physiological parameters of a human subject such as body temperature, blood oxygen level and heart rate. It can also detect a fall. The third part of the system is a web-enabled, robot-based vision system. The robot, operating in a home environment, can be remote controlled by a computer over the internet. The camera, mounted on the robot, can be used to send a picture of the affected area.
    Attached files: sensors and robotic environment for care of the elderly.pdf sensors and robotic environment for care of the elderly.ppt
    Automatic container-code recognition is of great importance to the modern container management system. Similar techniques have been proposed for vehicle license plate recognition in past decades. Compared with license plate recognition, automatic container-code recognition faces more challenges due to the severity of nonuniform illumination and invalidation of color information. In this paper, a computer vision based container-code recognition technique is proposed. The system consists of three function modules, namely location, isolation, and character recognition. In location module, we propose a text-line region location algorithm, which takes into account the characteristics of single character as well as the spatial relationship between successive characters. This module locates the text-line regions by using a horizontal high-pass filter and scanline analysis. To resolve nonuniform illumination, a twostep procedure is applied to segment container-code characters, and a projection process is adopted to isolate characters in the isolation module. In character recognition module, the character recognition is achieved by classifying the extracted features, which represent the character image, with trained support vector machines (SVMs). The experimental results demonstrate the efficiency and effectiveness of the proposed technique for practical usage.
    Attached files: An automated vision system for container-code recognition.pdf
    In this paper, by analyzing the number of concave domain and cycles and the characteristic of connected domain of the characters, we present a method of character recognition based on above mentioned features. At first, the characters are divided into two classes according to cycles: the character with cycles and the character without cycles. Then, we detect if the character has sunk parts in its four directions (up, down, left, right).According to the results, the character can be recognized or divided into six classes. At last, we recognize the character using the characteristic of connected domain, and obtain the final recognition results. The experiment shows that the algorithm has a good performance in English characters recognition.
    Attached files: A method of character recognition based on general characteristic and connected (2).pdf
    The recognition and tracking of traffic lights for intelligent vehicles based on a vehicle-mounted camera are studied in this paper. The candidate region of the traffic light is extracted using the threshold segmentation method and the morphological operation. Then, the recognition algorithm of the traffic light based on machine learning is employed. To avoid false negatives and tracking loss, the target tracking algorithm CAMSHIFT (Continuously Adaptive Mean Shift), which uses the color histogram as the target model, is adopted. In addition to traffic signal pre-processing and the recognition method of learning, the initialization problem of the search window of CAMSHIFT algorithm is resolved. Moreover, the window setting method is used to shorten the processing time of the global HSV color space conversion. The real vehicle experiments validate the performance of the presented approach.
    Attached files: The Recognition and Tracking of Traffic Lights Based on Color Segmentation.pdf
    Considering the problem of recognizing human actions from still images. We propose a novel approach that treats the pose of the person in the image as latent variables that will help with recognition. Different from other work that learns separate systems for pose estimation and action recognition, then combines them in an ad-hoc fashion, our system is trained in an integrated fashion that jointly considers poses and actions. Our learning objective is designed to directly exploit the pose information for action recognition. Our experimental results demonstrate that by inferring the latent poses, we can improve the final action recognition results.
    Attached files: Recognizing Human Actions from Still Images with Latent Poses CVRP2010.pdf Seminar-05-11-2011.pptx
    LPP(Locality Preserving Projection), as a linear version of manifold learning algorithm, has attracted considerable interests in recent years. For real time applications, the response time is required to be as short as possible. In this paper, a new Local image descriptor—LPP-HOG (Histograms of Oriented Gradients) for fast human detection is presented. We employ HOG features extracted from all locations of a grid on the image as candidates of the feature vectors. LPP is applied to these HOG feature vectors to obtain the low dimensional LPPHOG vectors. The selected LPP-HOG feature vectors are used as an input of linear SVM to classify the given input into pedestrian/non-pedestrian. We also present results showing that using these descriptors in human detection application results in increased accuracy and faster matching.
    Attached files: HoG.ppt LPP-HOG A New Local Image Descriptor for Fast Human Detection.pdf Histograms of Oriented Gradients for Human Detection.pdf
    Considering the problem of recognizing human actions from still images. We propose a novel approach that treats the pose of the person in the image as latent variables that will help with recognition. Different from other work that learns separate systems for pose estimation and action recognition, then combines them in an ad-hoc fashion, our system is trained in an integrated fashion that jointly considers poses and actions. Our learning objective is designed to directly exploit the pose information for action recognition. Our experimental results demonstrate that by inferring the latent poses, we can improve the final action recognition results.
    Attached files: Recognizing Human Actions from Still Images with Latent Poses CVRP2010.pdf
    Abstract—Overlay text brings important semantic clues in video content analysis such as video information retrieval and summarization, since the content of the scene or the editor’s intention can be well represented by using inserted text. Most of the previous approaches to extracting overlay text from videos are based on low-level features, such as edge, color, and texture information. However, existing methods experience difficulties in handling texts with various contrasts or inserted in a complex background. In this paper, we propose a novel framework to detect and extract the overlay text from the video scene. Based on our observation that there exist transient colors between inserted text and its adjacent background, a transition map is first generated. Then candidate regions are extracted by a reshaping method and the overlay text regions are determined based on the occurrence of overlay text in each candidate. The detected overlay text regions are localized accurately using the projection of overlay text pixels in the transition map and the text extraction is finally conducted. The proposed method is robust to different character size, position, contrast, and color. It is also language independent. Overlay text region update between frames is also employed to reduce the processing time. Experiments are performed on diverse videos to confirm the efficiency of the proposed method.
    Attached files: A new approach for overlay text detection and extaraction from complex video scene.pdf
    Abstract—The objective of this paper is to develop a robust and fast algorithm for vision-based road boundary detection.This paper proposes a flexible scenario to integrate two algorithms developed by our previous work for improving the precision and robustness of the lane boundary detection, and applied on the vision-based automated guided vehicle (AGV) system. In our previous study on vision-based AGV, the road boundary detection was used to measure the attitude of vehicle in order to guide along the lane center and keeps correct attitude. The traditional edge detection methods were being substituted for the histogram-based color difference fuzzy cluster analysis (HCDFCM) to fast recognize the lane boundary. Although HCDFCM held faster and more precise features than traditional methods, the shadowy road interfered in the precision of lane boundary detection. In this paper, we use fuzzy inference system (FIS) to enhance the contrast of shadowy pixels, and find the similarity with the lane model to solve the fault of detection problem in the case of shadowy situation. For the sake of reducing computational times adaptively, the enhanced algorithm provides a scene for incorporating HCDFCM with shadow removing algorithm. If the lane center variation on the image plane is larger than a certain threshold initialized by HCDFCM, the adjustable scan region on image plane uses to reinforce the robustness of lane boundary detection. The proposed method developed a feasible way to detect the lane boundary with high quality and reduced computational times.
    Attached files: the robust and fast approach for vision-based shadowy road boundary detection.pdf
    This paper proposes a novel image-processing algorithm to recognize the lane-curve of a structured road. The proposed algorithm uses an lane-curve function (LCF) obtained by the transformation of the defined parabolic function on the world coordinates into the image coordinates. Unlike other existing methods, this algorithm needs no transformation of the image pixels into the world coordinates. The main idea of the algorithm is to search for the bestdescribed LCF of the lane-curve on an image. In advance, several LCFs are assumed by changing the curvature. Then, the comparison is carried out between the slope of an assumed LCF and the phase angle of the edge pixels in the lane region of interest constructed by the assumed LCF. The LCF with the minimum difference in the comparison becomes the true LCF corresponding to the lane-curve. The proposed method is proved to be efficient through experiments for the various kinds of images, providing the reliable curve direction and the valid curvature compared to the actual road.
    Attached files: A lane-curve detection based on an LCF.pdf
    This paper presents a roust laser scan matching algorithm in dynamic environments. Scan matching is thought to be an essential function for mapping and localization of mobile robots. Out method is based on the random sample and Consensus (RANSAC) algorithm known for its good robust parameter estimation of the model parameters. Different from the existing scan matching methods for mobile robots, we only use the raw data of laser scanning without odometer information to find the transformation between two given laser data sets. Our method does not require any feature extraction and also need not initial estimation to reach global optimum. We demonstrate the practical usability of the proposed approach through Experiment.
    Attached files: Robust Laser Scan Matching in Dynamic Environments.pdf
    In this paper, we discuss how we use variances of gray level spatial dependencies as textural features to retrieve images having some section in them that is like the user input image. Gray level co-occurrence matrices at ¯ve distances and four orientations are computed to measure texture which is de¯ned as being speci¯ed by the statistical distribution of the spatial relationships of gray level properties. A likelihood ratio classi¯er and a nearest neighbor classi¯er are used to assign two images to the relevance class if they are similar and to the irrelevance class if they are not. A protocol that involves translating a K £ K frame throughout every image to automatically construct groundtruth image pairs is pro- posed and performance of the algorithm is evaluated accordingly. From experiments on a database of 300 512£512 grayscale images with 9,600 groundtruth image pairs, we were able to estimate a lower bound of 80% correct classi¯cation rate of assigning sub-image pairs we were sure were relevant, to the relevance class. We also argue that some of the assign- ments which we counted as incorrect are not in fact incorrect.
    Attached files: Content-based Image Database Retrieval Using Variances of Gray Level Spatial Dependencies.pdf
    Color image segmentation is an important but still open problem in image processing. In this paper, we propose a method for this problem by integrating the spatial connectivity and color feature of the pixels. Considering that an image can be regarded as a dataset in which each pixel has a spatial location and a color value, color image segmentation can be obtained by clustering these pixels into different groups of coherent spatial connectivity and color. To discover the spatial connectivity of the pixels, density-based clustering is employed, which is an effective clustering method used in data mining for discovering spatial databases. Color similarity of the pixels is measured in Munsell (HVC) color space whose perceptual uniformity ensures the color change in the segmented regions is smooth in terms of human perception. Experimental results using proposed
    Abstract¡ªThis paper demonstrates the use of the epipolar geometry to estimate the depth(distance) to a point on the object structure in a pair of stereo images. The well established robust method of RANSAC was implemented to calculate the fundamental matrix. The paper discusses how to calculate the disparity, a parallax that exists in the stereo images. This paper is generally in the nature of following up the existing technology, but addresses the stability issue of the RANSAC fundamental matrix algorithm as a stepping stone for stable generation of the dense disparity maps useful for applications.
    Abstract - In this paper, we present a robust method for detecting other vehicles from a forward-looking CCD camera in a moving vehicle. This system uses edge and shape information to detect other vehicles. The algorithm consists of three steps: lane detection, vehicle candidate generation, and vehicle verification. First, after detecting a lane from the template matching method, we divide the road into three parts: left lane, front lane, and right lane. Second, we set the region of interest (ROI) using the lane position information and extract a vehicle candidate from the ROI. Third, we use local orientation coding (LOC) edge image of the vehicle candidate as input to a pretrained neural network for vehicle recognition. Experimental results from highway scenes show the robustness and effectiveness of this method.
    This paper presents a tobust vehicle detection approach using Haar-like feature. It is possible to get a strong edge feature from this Haar-like feature. Therefore it is very effetive to remove the shadow of a vehicle on the road. And we can detect the boundary of vehicles accurately. In the paper, the vehicle detection algorithm can be divided into two main steps. One is hypothesis generaion, and the other is hypothesis verification. And in the second step, it determines whether the candidate is a vehicle or not by using the symmetry of vehicle edge features. In this research, we can get the detection rate over 15frame per second on our embedded system.
    Attached files: ____vehicel detection method using harr-like feature on real time system.pdf
    The vast majority of Autonomous Ground Vehicle in development today operate with GPS based navigation systems. While the accuracy of GPS systems has improved greatly over the previous decade, the stability of their signal has not. The phenomenon is commonly known as drift and may have a magnitude of more than a few meters. This paper outlines a method for vision based correction and localization of vehicle position through consideration of a priori information and perceived road characteristics. The approach is called Vision Based Position Correction for Urban Enviroments, and it will be deployed in the 2007 DARPA Urban Challenge
    In this paper, we present an algorithm that generates high dynamic range (HDR) images from multi-exposed low dynamic range (LDR) stereo images. The vast majority of cameras in the market only capture a limited dynamic range of a scene. Our algorithm first computes the disparity map between the stereo images. The disparity map is used to compute the camera response function which in turn results in the scene radiance maps. A refinement step for the disparity map is then applied to eliminate edge artifacts in the final HDR image. Existing methods generate HDR images of good quality for still or slow motion scenes, but give defects when the motion is fast. Our algorithm can deal with images taken during fast motion scenes and tolerate saturation and radiometric changes better than other stereo matching algorithms.
    Speckle noise of Synthetic aperture radar (SAR) image and the azimuth sensitivity for imaging obstruct the interpretation and applications for SAR images. But it is very much remote sensing source because it can acquire data under all-day and all weather. Texture features reflect the spatial property of ground objects. If texture features change, in general the ground object does so. Therefore, this paper deep analyzes SAR image texture feature and proposes a new multi-temporal SAR image change detection algorithm, which is called by texture features fusion voting (TFFV) algorithm. Finally, the airborne SAR image data tests the method and experimental results indicate that the presented methods are feasible.
    Attached files: A New Change Detection Algorithm for SAR Images.pdf
News | About us | Research | Lectures