Makito Seki

Human intruder detection with Leaky Coaxial cables for wide area surveillance system

We propose an algorithm to detect human intruders by means of Leaky Coaxial cables (LCX) for wide area surveillance systems. When a human intrudes into an area between two LCX, the radiated waves will suffer considerable changes with specific fluctuations. The former version of our detection algorithm had many false detections because the algorithm used only amplitudes of in-phase/quadrature-phase (I/Q) detector signals. To develop a sophisticated detection algorithm, we analyzes I/Q detector signals in detail and introduce some critical signal features appeared in I/Q detector signals to discriminate human intruders from background noises and other irrelevant objects. We successfully reduced many of the false detections compared to the performance of the former algorithm. In this paper we introduce our LCX-based surveillance system, and describe a human intruder detection algorithm based on I/Q detector signals with typical simulation results. Current problems and remaining tasks are also discussed.

Real-time violent action detector for elevator

This paper presents a new critical event detection method simplified for an embedded appliance mounted on an elevator car. We first define that the critical event is unusual action such as violent action, counteraction, etc, and introduce the violent action degree(VAD). We use an optical flow based method to analyze the current state of the motion through an ITV(Industrial TeleVision) camera. After motion analysis, we calculate a normalized statistical value, which is the VAD. The statistical value is the multiple of the optical flow direction variance, the optical flow magnitude variance, and optical flow area. Our method calculates the statistical value variance and normalize it by the variance. At last we can detect critical event by thresholding the VAD. Then we implement this method on an embedded appliance. The appliance has an A/D converter with special designed frame buffer, a 400MIPS high performance micro processor, dynamic memory, and some flash ROM. Since we need to process the method 4Hz or faster to keep the detection performance, we shrink the images into 80 by 60 size, adopt the recursive correlation method, and analyze optical flows. The special designed frame buffer enables us for capturing sequencial two images at any time. After that we achieve about 8Hz processing performance on it. Our method detects 80% of critical events where at most 6% of false acception.

Vehicle detection using Gaussian mixture model for infrared orientation-code image

This paper describes an approach to the detection of vehicles in infrared images. Stable vehicle detection is important for future intelligent transport systems and is generally done by background subtraction and object modeling. To avoid the daylight-dependent and weather-dependent influences of varying illumination in visible images acquired with conventional ITV cameras, some researchers have been using infrared (IR) images. IR images make it easy to extract foreground vehicle regions from background scenes, but their lack of clarity make object modeling difficult. We therefore propose a method that describes the internal pattern of each vehicle by using Gaussian mixture models (GMM) in the orientation-code image (OCI) space. Each pixel of an OCI has information about the maximum-gradient orientation of the IR image, not intensity information. Gradient orientation information does not depend on contrast and can describe the internal pattern structures of objects even in unclear IR images. We use the GMM to describe the topological structures of the internal patterns of vehicles. This approach can also eliminate the influences due to small differences between patterns. Evaluation tests with actual infrared video sequences have proved that the proposed algorithm provides stable vehicle detection.

Reliable image matching using binarized gradient features obtained with multi-flash camera

In this paper, we propose an object detection method using features describing information about a concavoconvex shape of an object that are obtained by using a small camera that controls the illumination direction. A feature image containing information about the shape of the object is generated by integrating images obtained by turning on, one by one, light emitting diodes (LEDs) annularly arranged around the camera. Our method can reliably detect a texture-less object by using this feature image in the matching process. Experiments using 200 actual images confirmed that the method achieves a 97.5% recognition success rate and a 4.62 sec processing time.

Robust Face Recognition System Using a Reliability Feedback

In the real world there are a variety of lighting conditions, and there exist many directional lights as well as ambient lights. These directional lights cause partial dark and bright regions on faces. Even if auto exposure mode of cameras is used, those uneven pixel intensities are left, and in some cases saturated pixels and black pixels appear. In this paper we propose robust face recognition system using a reliability feedback. The system evaluates the reliability of the input face image using prior distributions of each recognition feature, and if the reliability of the image is not enough for face recognition, it capture multiple images by changing exposure parameters of cameras based on the analysis of saturated pixels and black pixels. As a result the system can cumulates similarity scores of enough amounts of reliable recognition features from multiple face images. By evaluating the system in an office environment, we can achieve three times better EER than the system only with auto exposure control.