Kenichi Yamada

Understanding driving situations using a network model

In the rear-end collision avoidance system (RCAS), which incorporates the functions of detecting and reporting distance headway as well as of collision avoidance through the use of automatic braking, the development of the technique to integrate the advanced characteristic of each sensors becomes important. In this paper, we propose the uniting processing technique which uses the network as a trial. We classify the recognition system into some modules from the software point of view, link these modules as the network and make the states of the network correspond to the driving environment. We referred to a immune network when making the network. The immune network is a technological model of the immune mechanism in the living body. We applied this network mechanism to fuse functions of the distance headway, the road lane position and the position of the preceding vehicle obtained by the laser radar and image processing. These three functions are transferred to modules, each module tests mutually, and each certainty degree is changed continuously. This paper describes this network technique and the experimental results.

Vehicles detection using sensor fusion

Sensor fusion method is more robust than the method using a single sensor, so sensor fusion is effective for vehicle recognition in a complex scene. However, if each sensing data is processed individually for most of the stages, recognition performance is not always good. In this paper, we propose the extensible and generalized fusion method. First, fusion vector which is combined with image sensor data and laser radar data at a primitive level is prepared. We regard fusion vector as sensing data by one robust sensor. Next, fusion vector is compared with a discriminated dictionary. We report the efficiency of our method at a complex scene in which recognition error tends to occur by using a single sensor.

Preceding vehicle and road lanes recognition methods for RCAS using vision system

This paper describes the preceding vehicle and road lanes recognition methods for the rear-end collision avoidance system (RCAS) which we are developing. These methods are using an edge histogram method based on the model based vision concept. The edge histogram method can detect line elements of the objects stably with low calculation cost. When the region of interests for the preceding vehicle and road lanes in the image are established and their projected edge histograms are observed in time series order, we can recognize them. Furthermore, we apply Kalman filter to their motions and predict their locations for time series detection. Using this stable recognition, we derive a collision time to control the on-board brake system. We show the performance of these methods by experimental results.

Image understanding based on edge histogram method for rear-end collision avoidance system

To avoid rear-end collision with the preceding vehicles on the road, we need the information of collision potentiality. It is decided primarily by the relative locations and velocities between vehicles and their absolute locations on the road lanes. We classify the recognition objects in the road scene into road lanes and vehicles for the goal of collision avoidance. We propose the consistent road scene recognition method using edge histogram with model based vision. The edge histogram can detect line elements of the objects stably with low calculation cost. If the suitable region of interests for each objects in the model are established and their projected edge histograms are observed in time series order, we can derive each objects from the model. Furthermore, we apply Kalman filter to predict the object locations for time series detection. From the recognition results, we can calculate the collision time. It is one of the measures of collision potentiality.<<ETX>>

Preceding Vehicle Recognition Algorithm Using Fusion Of Laser Radar And Image Processing

This paper describes a fusion system combining the properties of laser radar and image processing. The system recognizes preceding vehicle positions and measures distance data using the laser radar, and improves certainty of the recognition results. The paper presents a general structure of this fusion system, and algorithms for recognizing and chasing vehicles. Details about original fast tracking algorithms are described.

Vehicle Detection Based on Perspective Transformation Using Rear-View Camera

In recent times, the number of vehicles with a rear-view camera has been increasing. The rear-view camera can be utilized as a sensor for monitoring a collision from behind the vehicle in a driving scene. To prevent rear-end collisions, we have been developing a technology that detects approaching vehicles from images obtained using an onboard rear-view camera. In conventional vehicle detection methods, often, camera-view images are used. However, it is difficult to accurately estimate the position of a distant approaching vehicle using such images. In this paper, we propose an improved method to accurately estimate the position of distant approaching vehicles by using virtual top-view images. The displacement of the vehicle in the top-view image is proportional to its speed. Thus, the proposed method can provide the accurate position of the distant vehicle. We describe the details of the proposed method and its availability by the experiment using actual images.

Road lane recognition system for RCAS

Road-lane recognition is one of the key technologies utilized in the intelligent transport system, a sophisticated road traffic information system in Japan. In this paper, we take a general view of road-lane recognition methods that use image processing with reference to the recognition frames of capturing, conversion, classification and interpretation. While these recognition methods employ a bottom-up processing approach, we believe a top-down processing approach based on prior knowledge of the recognition target will play a more important role in the future. One of the purposes of road-lane recognition in rear-end collision avoidance system (RCAS) is to determine whether the vehicle in front is in the same lane or not. As a solution to that question, we introduce a network type fusion method, which divides a recognition process into modules connected in a network and then uses the changes of state obtained in mutual tests to determine the vehicles position relative to the road lane.

The omnidirectional vision sensor for in-vehicle image processing applications

The omnidirectional vision sensor can view the environments around the vehicle. Typical type of this sensor is the fish-eye typed and the convex mirror typed. In this paper, we described approaches to recognize surrounding moving objects with the convex mirror typed omnidirectional vision sensor As a example of recognition method of the vehicle, we tried to observe the radial projection about intensity of the deferential image in time series order. Furthermore, we tried to give surrounding monitoring system as the proper application to the vehicle with the convex mirror type.