Hiroshi Hanaizumi

Barcode readers using the camera device in mobile phones

This paper shows new algorithms and the implementations of image reorganization for EAN/QR barcodes in mobile phones. The mobile phone system used here consists of a camera, mobile application processor, digital signal processor (DSP), and display device, and the source image is captured by the embedded camera device. The introduced algorithm is based on the code area found by four corners detection for 2D barcode and spiral scanning for 1D barcode using the embedded DSP. This algorithm is robust for practical situations and the DSP has good enough performance for the real-time recognition of the codes. The performance of our image processing is 66.7 frames/sec for EAN code and 14.1 frames/sec for QR code image processing, and this is sufficient performance for practical use. The released mobile phone had performance of 5-10 frames/sec including OS and subsystem overheads.

A change detection method for remotely sensed multispectral and multitemporal images using 3-D segmentation

A new method is proposed fur detection of the temporal changes using three-dimensional (3D) segmentation. The method is a kind of clustering methods for temporal changes. In the method, multitemporal images form a image block in 3D space; x-y plane and time axis. The image block is first divided into spatially uniform sub-blocks by applying binary division process. The division rule is based on the statistical t-test using Mahalanobis distance between spatial coefficient vectors of a local regression model fitted to neighboring sub-blocks to be divided. The divided sub-blocks are then merged into clusters using a clustering technique. The block-based processing, like the spatial segmentation technique, is very effective in reduction of apparent changes due to noise. Temporal change is detected as a boundary perpendicular to the time axis in the segmentation result. The proposed method is successfully applied to actual multitemporal and multispectral LANDSAT/TM images.

Change Detection From Remotely Sensed Multi-temporal Images Using Spatial Segmentation

A new method is proposed for detection of temporal change between remotely sensed multi-spectral images. This method uses regression and segmentation techniques for reduction of apparent change due to noise and/or variation of recording gain and offset between two images. In the method, the images are normalized by fitting a line to the whole pixel scattergram between them. The line is specified by slope and intercept. Temporal change is detected when the hypothesis is rejected that a line fitted to the whole pixel scattergram is identical to that to a local area scattergram. The hypothesis is statistically tested. Noise is reduced by above regression process. The spatial segmentation is used for division of the image into small areas where density scattergrams are described well by a single set of slope and intercept. In this paper, we describe the principle and the procedures of this method. The validity of this method is confirmed by numerical simulation. A result obtained by application of this method to a pair of actual multi-temporal multi-spectral images is also shown.

A flexible method for recognizing four-digit numbers on a license-plate in a video scene

In order to realize a simple system for monitoring car movements, a flexible method was developed for recognizing four-digit numbers on a license-plate in a video scene. The method consisted of three parts; global search for finding a license plate in a scene, corner detection for removing shape deformation, and numeral character recognition. The spatial features and the spectral (color) ones were merged and processed at a time in the global search. The Hough transformation was adopted for selecting sides of the license-plate among various edges in the following corner detection. The corner points found as cross points of the neighboring sides were used for removing the shape deformation. The four-digit numbers were recognized by using a virtual pixel algorithm with centroid compensation. The global search achieved 97 % of correct detection rate for 100 sets of scenes. The corner detection achieved 65 % (39 of 65 scenes) of correct detection rate, so far. The four-digit numbers on the license-plates, whose corners were correctly detected, were recognized perfectly.

Depthmap-based obstacle avoidance on rough terrain

It is an important and difficult task for a mobile robot to detect the maneuverable area and to avoid obstacles. Numerous methods using ultra sonic or infrared sensors have been successfully proposed and adapted to various robots. These robots, however, cannot move around in unleveled areas because their sensors are incapable of measuring in a three-dimensional (3D) environment. Some obstacle avoidance methods using 3D sensors were proposed. It is a problem to move the robot around smoothly and continuously, considering that these methods require numerous processing steps and time to determine the obstacles and the avoidance path. In this paper, a new obstacle avoidance method using a depthmap is proposed that applies 3D sensor technology to autonomous exterior robot systems. Employing this method the robot is able to recognize the obstacle continuously and thus can be used on rough terrain. All processes in the algorithm handle images without the environmental model. In recent years, a sensing device used for laser modules or CCD cameras that can sense 3D space in real-time have been developed. Using such a sensor in conjunction with the proposed method, the test unit and autonomous vehicle were built to evaluate the methodology in an outdoor environment These units were able to detect the obstacle, avoid it, and Iocomote autonomously on rough terrain. As a result, this method has the advantage of robustness on rough terrain, and an autonomous mobile robot with this test unit is capable of navigation on an unleveled road.

A change detection method for remotely sensed multi-spectral and multi-temporal images using 3-D segmentation

A new method is proposed for detection of the temporal changes using 3-dimensional segmentation. The method is regarded as a clustering method for temporal changes. The spatial segmentation technique is used for reduction of apparent change due to noise. The proposed method is successfully applied to actual multi-temporal and multispectral LANDSAT/TM images.

A binary division algorithm for clustering remotely sensed multi-spectral images

A new method is proposed for clustering remotely sensed multi-spectral images with both high accuracy and high efficiency. For high speed processing, we project image data onto one dimensional sub-space, and limit the number of boundaries in the sub-space. The optimal sub-space and boundary are selected so that the ratio of the variance of within distance to the variance of between distance takes the minimum value. Image data are repeatedly divided into two groups until all of the groups consist of a single cluster. Performance of the proposed method was better than that of ISODATA in both speed and accuracy. The method was successfully applied to actual remotely sensed multi-spectral images. >

An automated method for understanding road traffic signs in a video scene captured by a mobile camera

We have already proposed a method for detecting road traffic signs in a video scene. In order to realize flexible detection in shape deformation due to discrepancy between target position and camera direction, multiple template techniques were introduced. Here, we proposed an automated method for understanding road traffic signs in a video scene. The method was located at the 2nd process for understanding the signs detected by the previous method. The sign understanding process was reduced to evaluation of both spatial and spectral similarities among the sign templates with possible deformations and signs detected. A binary decision tree classifier was introduced for efficient performing the evaluations.

An automated method for estimating terrain height from SPOT stereo pair images

An automated method is proposed for producing a digital terrain model from SPOT/HRV stereo pair images. The method provides terrain height by using absolute orientation with the minimum number (two) of ground control points. The precision of terrain height obtained from an actual stereo pair image was estimated to be within 20 m.

Method for change analysis with weight of significance using multitemporal multispectral images

A new method is proposed for change analysis with weight of significance between two multi- temporal multi-spectral images. This method gives us areas which indicate the assigned temporal change, for example, from vegetation to bare soil. Image data are projected onto a feature space in which the assigned change is emphasized, and temporal changes between two images are detected with suppression of irrelevant changes. The validity of the method is confirmed by numerical simulation. The method is successfully applied to actual multi- temporal and multi-spectral images.