Naoya Ohta

A statistical approach to background subtraction for surveillance systems

Background subtraction is a commonly used process in surveillance systems. One difficult problem when using the process is maintaining a correct background image against changing illumination conditions. Most methods for maintaining the background image are based on intuitive definitions about the illumination change and are implemented as somewhat ad hoc algorithms. In contrast, we first define mathematical models representing the relation between the illumination intensity, a reflection index of objects and a pixel value. We also mathematically define an assumption about illumination, which requires that the distribution of the illumination intensity in a small region does not change. Then we formalize the background subtraction problem as a statistical test (/spl chi//sup 2/ test) based on the models and assumption. The experiments show that our models appropriately express the imaging process of a camera and our method provides stable detection performance for foreground objects.

Accuracy bounds and optimal computation of homography for image mosaicing applications

We describe a theoretically optimal algorithm for computing the homography between two images in relation to image mosaicing applications. First, we derive a theoretical accuracy bound based on a mathematical model of image noise and do simulation to confirm that our renormalization technique effectively attains that bound; our algorithm is optimal in that sense. Then, we apply our technique to mosaicing of images with small overlaps. By using real images, we show how our algorithm reduces the instability of the image mapping.

Optimal Estimation of Three-Dimensional Rotation and Reliability Evaluation

We discuss optimal rotation estimation from two sets of 3-D points in the presence of anisotropic and inhomogeneous noise. We first present a theoretical accuracy bound and then give a method that attains that bound, which can be viewed as describing the reliability of the solution. We also show that an efficient computational scheme can be obtained by using quaternions and applying renormalization. Using real stereo images for 3-D reconstruction, we demonstrate that our method is superior to the least-squares method and confirm the theoretical predictions of our theory by applying the bootstrap procedure.

Modeling Surperspective Projection of Landscapes for Geographical Guide-Map Generation

It is still challenging to generate hand‐drawn pictures because they differ from ordinary photographs in that they are often drawn as seen from multiple viewpoints. This paper presents a new approach for modeling such surperspective projection based on shape deformation techniques. Specifically, surperspective landscape images for guide‐maps are generated from 3D geographical elevation data. Our method first partitions a target geographical surface into feature areas to provide designers with landmarks suitable for editing. The system takes as input 2D visual effects, which are converted to 3D geometric constraints for geographical surface deformation. Using ordinary perspective projection, the deformed shape is then transformed into a target guide‐map image where each landmark enjoys its own vista points. An algorithm for calculating such 2D visual effects semi‐automatically from the geographical shape features is also considered.

Creating a Worldwide Community Security Structure Using Individually Maintained Home Computers: The e-JIKEI Network Project

This article describes a project for creating a community security structure spread worldwide that uses individually maintained home computers connected to the Internet. The basic concept of the project is that every individual person watches in and around his or her house using cheap cameras as his or her eyes, personal computer as his or her brain, and Internet as his or her communication means by his or her own expense. The purpose of the project is to re-create, in the present day, a mutual watch system, which widely existed in old Japanese communities, as well as in other countries in the world, but in a much more powerful and flexible form with the aid of the information technology. In the first stage of the project, a preliminary system with the minimum necessary functions for realizing the concept has been developed, which consisted of free software and operation manuals.

Smart street light system looking like usual street lights based on sensor networks

Currently, in the whole world, enormous electric energy is consumed by the street lights, which are automatically turn on when it becomes dark and automatically turn off when it becomes bright. This is the huge waste of energy in the whole world and should be changed. This paper discusses a smart street light system, whose concept is proposed by Fujii et al. The main aim of smart street light systems is that lights turn on when needed and light turn off when not needed. Moreover, the smart street light system in this paper behaves like usual street lights that turn on all night. The ideal behavior of the smart street light system is that no one finds turn-off of street lights at night. Whenever someone see street lights, they turn on and whenever no one see street lights, they turn off. The smart street light system consists of LED lights, brightness sensors, motion sensors and short-distance communication networks. The lights turn on before pedestrians and vehicles come and turn off or reduce brightness when there is no one. It will be difficult for pedestrians and drivers of vehicles to distinguish our smart street lights and the conventional street lights because our street lights all turn on before they come.

Comparing optimal three-dimensional reconstruction for finite motion and optical flow

We present two linear algorithms for 3-D reconstruction: one is for finite motion; the other is for optical flow. We compare their performance by simulation and real-image experiments, using the same data. The two algorithms are both theoretically optimal, ex- tracting maximum information from the input. We observe that the 3-D reconstruction by the finite motion algorithm is generally more accurate than by the optical flow algorithm if point correspondences are identified with optical flow.

Optical Flow Detection Using a General Noise Model for Gradient Constraint

In the usual optical flow detection, the gradient constraint, which expresses the relationship between the gradient of the image intensity and its motion, is combined with the least-squares criterion. From a statistical point of view, this means assuming that only the time derivative of the image intensity contains Gaussian noise. However, it is more reasonable to assume that all the derivatives are observed with Gaussian noise because they are equally computed from pixels containing noise and approximated by finite difference. In this paper, we study a new optical flow detection method based on the latter assumption. Since this method requires the knowledge about the covariance matrix of the noise, we also discuss a method for its estimation. Our experiments show that the proposed method can compute optical flow more accurately than the conventional method.

Smart street light system with energy saving function based on the sensor network

Our project for developing a smart street light system is reviewed. In this project, the street light system, in which lights on when needed and light-off when not needed. Currently, in the whole world, enormous electric energy is consumed by the street lamps, which are automatically turn on when it becomes dark and automatically turn off when it becomes bright. This is the huge waste of energy in the whole world and should be changed. Our smart street light system consists of a LED light, a brightness sensor, a motion sensor and a short-distance communication network. The lights turn on before pedestrians and vehicles come and turn off or reduce power when there is no one. It will be difficult for pedestrians and drivers of vehicles to distinguish our smart street lamps and the conventional street lights, since our street lamps all turn on before they come. The present status and the future prospects of our smart start light project will be reviewed.

How Much Does Color Information Help Optical Flow Computation?*This work was in part supported by the Ministry of Education, Culture, Sports, Science and Technology, Japan under a Grant in Aid for Scientific Research C (2) (No.16500099).

Optical flow is usually computed only on the basis of intensity information of images. Therefore, if we use color information in addition to the intensity, it is expected that more accurate optical flow can be computed. However, this intuition will be correct only when the following conditions are satisfied. First, the images should contain rich color variations. Moreover, it is also required that the image gradient of each color band differs in its direction. In this report, we empirically examined the difference of gradient directions on each band using 500 images, and evaluated quantitatively the advantage of using color information for optical flow computation.