Abstract: A method includes generating a disparity map according to two sets of image data, identifying at least one object in the disparity map, mapping the at least one object onto a plane view, tracking at least one object on the plane view, and providing a robust algorithm about a cross-line time interval and a cross-line degree. The robust algorithm includes detecting whether the at least one object enters a predetermined region when the at least one object crosses a predetermined boundary.

Abstract: A method of performing video stabilization may be applied to unstable frames captured by a vibrating camera. The method includes checking if the frame is vibrating and calculating a degree of vibration for compensating a vibration area. Based on block motion estimation, a feature block selected from nine fixed observation blocks is used to obtain a vibration vector. The vibration vector is used to adjust the vibration frame for restoring the frame.

Abstract: A method for intensifying identification of three-dimensional objects identification includes utilizing a left eye camera and a right eye camera to capture a left eye image and a right eye image, calibrating the left eye image and the right eye image to generate a calibrated left eye image and a calibrated right eye image, using the calibrated left eye image and the calibrated right eye image to generate a disparity map, differentiating a three-dimensional object from a background image according to the disparity map, projecting the three-dimensional object onto a plan view, filtering noise out of the plan view to generate a filtered three-dimensional object, determining whether the filtered three-dimensional object contains at least two three-dimensional objects, and separating the at least two three-dimensional objects if the filtered three-dimensional object contains at least two three-dimensional objects.

Abstract: Determining if an input image is a foggy image includes determining an average luminance gray level of the input image, performing Sobel image processing on the input image to generate a Sobel image of the input image when the average luminance gray level of the input image is between a first image average luminance and a second image average luminance, determining a first normalization value and a second normalization value of the input image, determining a mean value and a standard deviation of the Sobel image when the first normalization value and the second normalization value are less than a first threshold value, and determining the input image as a foggy image when a sum of the mean value and the standard deviation is less than a second threshold value.

Abstract: While locating a license plate of a moving vehicle on consecutive images, motion detection is first performed on the consecutive images to detect a moving vehicle image, which is segmented using edge detection, and the segmented moving vehicle image is analyzed to retrieve characteristics for locating a license plate image and determining characters on the located license plate image. As a result, a precise location of the license plate is thus precisely located for further recognition no matter what weathers in which the consecutive images are recorded. The above-mentioned technique requires merely few calculations, is easily implemented, and may be applied on an intelligent digital video recording (DVR) system including many computer-vision functions.

Abstract: A method for reducing image noise includes calculating a first pixel amount of pixels that are similar to each other in a first number neighbor of a center pixel in a motion window, determining whether the first pixel amount of pixels that are similar to each other in the first number neighbor is greater than a first predetermined value, and using a mean of those pixels of the first pixel amount of pixels that are similar to each other in the first number neighbor to restore the center pixel of the motion window if the first pixel amount of pixels is greater than the first predetermined value. The method includes determining whether a second pixel amount of pixels that are similar to the center pixel is greater than a second predetermined value if the first pixel amount of pixels is not greater than the first predetermined value.

Abstract: Comparison of parameters, including width, length, depth, color, and shape of a target object with a reference object is performed through use of a stereo camera. If the parameters of the target object are within threshold values of the parameters of the reference object, a match is indicated. If not, a new reference object is selected for comparison with the target object.

Abstract: Vehicle segmentation and counting method based on the property of color variation and headlight information combining change detection in nighttime traffic environment (one-way road with coming direction) is provided. The goal is to reduce the effect of ground-illumination that decreases the accuracy of vehicle segmentation. Besides, the amount of traffic flow is calculated and it can be used in other post-applications, such as traffic-flow reporting or controlling.

Abstract: Disclosed is a multi-path method of block matching computing for image processing, which comprises: determining a searching pattern with a center pixel point and a plurality of surrounding pixel points to perform a block searching operation; computing a plurality of pixel value differences between the center pixel point and the surrounding points of two continuous frames; and utilizing at least two pixel points with a pixel value difference as the center pixel point to perform the next block searching operation, wherein the pixel difference of the at least two pixel points has a difference less than a predetermined value from a minimum pixel value difference.

Abstract: Disclosed is a method for video object segmentation, which includes the following steps:(a) detecting a changing state between a current frame and a previous frame; (b) categorizing the current frame as a moving region, a still region, a background region and an uncovered background region according to the detecting result from the step (a); (c) determining a moving object in the current frame according to the sorting result from the step (b); and (d) updating a background component.

Abstract: A video object segmentation method takes advantage of edge and color features in conjunction with edge detection and change detection to improve the accuracy of video object segmentation for rainy situations. The video object segmentation method of the present invention includes analyzing HSI-color information of the initially extracted objects to obtain features of the moving object; performing edge detection to obtain edges of the moving object for reducing the effect of confusing raindrops with moving objects in rainy dynamic background; performing object region detection to generate an accurate object mask for solving the uncovered-background problem and the still-object problem; and employing a bounding-box based matching method for solving the reflection problem of the moving object in the rained ground.

Abstract: A method to calculate a transform coordinate on a second video of an object having an target coordinate on a first video and related operation process and video surveillance system are disclosed. The method includes the steps of generating the first video with a first video surveillance device, generating the second video with a second video surveillance device, selecting a first predetermined number of first reference objects also existing on the second video from the first video, calculating a plurality of coordinate transform coefficients of a first coordinate transform function according to the first predetermined number of first coordinates where the first reference objects are on the first video and the first predetermined number of second coordinates where the first reference objects are on the second video, and substituting the target coordinate into the first coordinate transform function, which had the first coordinate transform coefficients, and calculating the transform coordinate.

Abstract: A method of effective image authentication and image restoration by hiding watermarks in DCT (Discrete Cosine Transform) coefficients is presented. The basic concept is to embed the selected significant watermarking bits for authentication and restoration into the selected medium- and low-frequency DCT coefficients. Thus, the illegally tampered regions can be detected, and then the original information in that region can be extracted for restoration. Experimental results show that the proposed authentication and restoration techniques can be applied to a DVR (Digital Video Recorder) system, in which no original image information is involved, and it can effectively detect the illegally tampered region and restore the tampered region in the human visual perceptual quality by only using a little embedded original information.

Abstract: Based on area and color analyses, a cost-effective bi-directional people counter dedicated to the pedestrian flow passing through a gate or a door is proposed. Firstly, the passing people are roughly counted with the area of people projected on an image captured by a zenithal video camera. The moving direction of the pedestrian can be recognized by tracking each people-pattern with an analysis of its HSI histogram. To improve the accuracy of counting, the color vector extracted from the quantized histograms of intensity or hue is introduced to refine the early counting. Besides, the inherent problems of both people touching together and merge/split phenomenon can be overcome.

Abstract: A method for reducing image noise includes calculating a mean pixel-correlation between a central pixel and neighboring pixels within a working window in an image. The method also includes calculating a weighted horizontal similarity between the central pixel and horizontal neighboring pixels and a weighted vertical similarity between the central pixel and vertical neighboring pixels within the working window. Then, the method determines if the central pixel is noise. If so, a noise-reduction process is executed.

Abstract: A method includes calculating a mean of a plurality of pixels of a motion window, calculating a pixel amount of pixels similar to a center pixel, calculating a variance of the pixels, determining whether a difference between the center pixel and the mean is greater than a first predetermined value, determining whether the pixel amount similar to the center pixel is greater than a second predetermined value if the difference between the center pixel and the mean is not greater than the first predetermined value, determining whether the variance is smaller than a threshold value if the pixel amount similar to the center pixel is not greater than the second predetermined value, and filtering the center pixel according to a result of determining whether the variance is smaller than the threshold value. Finally, temporal weighted mean filters involving motion estimation are used for motion compensation in images after spatial filtering.

Abstract: A flatted hexagon search method selects seven initial check points over three adjacent rows with six of the initial check points positioned at vertices of a hexagon and one of the initial check points positioned at a center of the hexagon. Then the seven initial check points are searched to find a minimum BDM (block distortion measure) point. With the flatted hexagon, the method will optimize speed and matching probability. Therefore the method is an attractive algorithm to estimate motion vectors for real world video data.

Abstract: Fire detection and smoke detection use an image capturing device to obtain images from a predetermined area and adopt an RGB (red, green, blue) color model based chromatic and disorder measurement for extracting flame pixels and smoke pixels. The extracted pixels are inputted to a fire detection fuzzy alarm system to generate an output of alarm information. Based on iterative checking on the growing ratio of the alarm information, a fire alarm is released accordingly.

Abstract: The static and dynamic characteristics of smoke are developed for extracting smoke pixels of an image and effective smoke detection. In static characteristic, a chromaticity-based analysis is employed to extract grayish pixels in the moving region from a monitored area. Then, a circumference-based disorder measurement in dynamic characteristic is developed to obtain the smoke pixels from the grayish pixels. To enhance the reliability of alarming, a fuzzy system is introduced to judge if the smoke will spread out by checking its growth-rate.

Abstract: Fire detection and smoke detection use an image capturing device to obtain images from a predetermined area and adopts an RGB (red, green, blue) color model based chromatic and disorder measurement for extracting fire pixels and smoke pixels. The extracted pixels are inputted to a fire detection fuzzy alarm system to generate an output of alarm information. Based on iterative checking on the growing ratio of the alarm information, a fire alarm is released accordingly.