Masatoshi Kimachi

A stereo-based vehicle detection method under windy conditions

We have developed a stereo-based video vehicle detector, Silhouette Vision/sup TM/ and its high ability of detecting vehicles has also been confirmed through various field tests. The main issue of the paper is to discuss vehicle detection of Silhouette Vision/sup TM/ under a windy condition. For effective vehicle detection, in general, a stereo-based video detector such as Silhouette Vision/sup TM/ requires accurate 3-D measurements in the roadway image. However, camera sway caused by the wind sometimes has a bad influence upon the calculation of 3-D measurements. To alleviate such a problem, we propose a 3-D measurements correction method that is effective even in windy conditions. In our simulation experiments, we show that Silhouette Vision/sup TM/ has enough ability to detect vehicles even under windy conditions.

Efficient contrast invariant stereo correspondence using dynamic programming with vertical constraint

In this paper, we propose a dense stereo algorithm based on the census transform and improved dynamic programming (DP). Traditional scanline-based DP algorithms are the most efficient ones among global algorithms, but are well-known to be affected by the streak effect. To solve this problem, we improve the traditional three-state DP algorithm by taking advantage of an extended version of sequential vertical consistency constraint. Using this method, we increase the accuracy of the disparity map greatly. Optimizations have been made so that the computational cost is only increased by about 20%, and the additional memory needed for the improvement is negligible. Experimental results show that our algorithm outperforms many state-of-the-art algorithms with similar efficiency on Middlebury College’s stereo Web site. Besides, the algorithm is robust enough for image pairs with utterly different contrasts by using of census transform as the basic match metric.

A robust 3d feature-based people stereo tracking algorithm

This paper presents a 3D feature-based people tracking algorithm which combines an interacting multiple model (IMM) algorithm with a cascade multiple feature data association algorithm. The IMM algorithm in this paper only uses an adaptive Kalman Filter and two dynamic models consisting of a constant velocity model (CV) and a current statistics model (CS) to predict the 3D location of people maneuvering and update the prediction with corresponding measurement. The cascade multiple feature data association algorithm in this paper utilizes three hypotheses, including the nearest distance hypothesis, the velocity consistency hypothesis, and the intensity consistency hypothesis, in turn to determine which trajectory a measurement should be assigned to. Experimental results demonstrate the robustness and efficiency of the proposed framework. It is real-time and not sensitive to the variable frame to frame interval time. It also can deal with the occlusion of people and do well in those cases that people rotate and wriggle.

Application of fuzzy theory to handwritten character recognition

There has been much research in recent decades on character recognition methods, and some methods have already been put into practical use. There are many unresolved problems, however, with respect to handwritten character recognition as composed with printed character recognition. The authors considered discriminant functions, which constitute the most important part of a character recognition method. As a result of considering problems of conventional statistical discriminant functions, the authors propose applying the fuzzy theory to discriminant functions. The so-called fuzzy discriminant function is capable of representing a data distribution in a more flexible manner because it consists of membership functions on the principal axes of learning samples. The authors conducted recognition experiments for handwritten characters with two types of membership functions. In one type the membership values are directly tuned based on human experiences; in the other they are derived from histograms or statistical data. With the former membership function, the recognition rate of 99.0 percent is achieved for [numeric] characters from the handwritten alphanumeric data base ETL6, and with the latter, the rate of 96.0 percent for [hiragana] characters from handwritten educational [kanji] data base ETL8. This result proves the effectiveness of the fuzzy discriminant function. It also indicates that a dynamic combination of human experiences and statistical techniques is a key to practical systems.

Shape Recovery of Specular Surface Using Color Highlight Stripe and Light Source Coding

Shape recovery of specular surface is a challenging task; camera images of these surfaces are difficult to interpret because they are often characterized by highlights. Structured Highlight approach is a classic and effective way for specular inspection, this paper suggests a new strategy to recover dense normals of a specular surface and reconstruct its shape by combining the ideas of Structured Highlight, color source coding, highlight stripe and its translations. Point sources with different colors are positioned on orbits to illuminate a specular object surface. These point sources are scanned, and highlights on the object surface resulting from each point source are used to derive local surface orientation. Dense normal information can be recovered by translating these orbits. Some experimental system configurations are given. The simulation results show that the new method is feasible and can be used to reconstruct shape of specular surface in a high precision.

A 3D Feature-Based Binocular Tracking Algorithm*Our work is supported by the National Natural Science Founds of China (No. 60271033) and SVS 04 of OMRON Corporation, Japan.

This paper presents a 3D feature-based binocular tracking algorithm for tracking crowded people indoors. The algorithm consists of a two stage 3D feature points grouping method and a robust 3D feature-based tracking method. The two stage 3D feature points grouping method can use kernel-based ISODATA method to detect people accurately even though the part or almost full occlusion occurs among people in surveillance area. The robust 3D feature-based Tracking method combines interacting multiple model (IMM) method with a cascade multiple feature data association method. The robust 3D feature-based tracking method not only manages the generation and disappearance of a trajectory, but also can deal with the interaction of people and track people maneuvering. Experimental results demonstrate the robustness and efficiency of the proposed framework. It is real-time and not sensitive to the variable frame to frame interval time. It also can deal with the occlusion of people and do well in those cases that people rotate and wriggle.

Robust estimation of less featured road surface for stereo-based vehicle detection

The conventional video vehicle detectors have some problems such as difficulty of parameter adjustment at initial settings that influences the detection performance, in addition to the performance degradation in adverse conditions. To solve these problems, stereo-based video vehicle detectors have been proposed. We propose a novel method to estimate relations automatically between a stereo camera and a road surface utilizing vehicles moving on the road.