Kiyosumi Kidono

Autonomous visual navigation of a mobile robot using a human-guided experience

Information on the surrounding environment is necessary for a robot to move autonomously. Many previous robots use a given map and landmarks. Making such a map is, however, a tedious work for the user. Therefore this paper proposes a navigation strategy which requires minimum user assistance. In this strategy, the user first guides a mobile robot to a destination by remote control. During this movement, the robot observes the surrounding environment to make a map. Once the map is generated, the robot computes and follows the shortest path to the destination autonomously. To realize this navigation strategy, we develop: (1) a method of map generation by integrating multiple observation results considering the uncertainties in observation and motion, (2) a fast robot localization method which does not use explicit feature correspondence, and (3) a method of selecting effective viewing directions using the history of observation during the guided movement. Experimental results using a real robot show the feasibility of the proposed strategy.

Experimental on Hierarchical Transmission Scheme for Visible Light Communication using LED Traffic Light and High-Speed Camera

LEDs are expected as lighting sources for next generation, and data transmission system using LEDs attract attention. In this paper, we present hierarchical coding scheme using LED traffic lights and high-speed camera for intelligent transport systems (ITS) application. Further, if each of LEDs in traffic lights is individually modulated, parallel data transmissions are possible using a camera as a reception device. Such parallel LED-camera channel can be modeled as spatial low-pass filtered channel of which the cut-off frequency varies according to the distance. To overcome, we propose hierarchical coding scheme based on 2D fast Haar wavelet transform. As results, the proposed hierarchical transmission schemes outperform the conventional on-off keying and the reception of high priority data is guaranteed even LED-camera distance is further.

Pedestrian recognition using high-definition LIDAR

Pedestrian detection is one of the key technologies for autonomous driving systems and driving assistance systems. To predict the possibility of a future collision, these systems have to accurately recognize pedestrians as far away as possible. Moreover, the function to detect not only people walking but also people who are standing near the road is also required. This paper proposes a method for recognizing pedestrians by using a high-definition LIDAR. Two novel features are introduced to improve the classification performance. One is the slice feature, which represents the profile of a human body by widths at the different height levels. The other is the distribution of the reflection intensities of points measured on the target. This feature can contribute to the pedestrian identification because each substance has its own unique reflection characteristics in the near-infrared region of the laser beam. Our approach applies a support vector machine (SVM) to train a classifier from these features. The classifier discriminates the clusters of the laser range data that are the pedestrian candidates, generated by pre-processing. A quantitative evaluation in a road environment confirms the effectiveness of the proposed method.

Pedestrian detection and direction estimation by cascade detector with multi-classifiers utilizing feature interaction descriptor

This paper proposes a pedestrian detection and direction estimation method by the cascade approach with multiclassifiers using the Feature Interaction Descriptor (FIND). FIND describes the high-level properties of an objects appearance by computing pair-wise interactions of adjacent regionlevel features. To perform efficient and accurate detection using FIND, we employ the cascade approach with multiclassifiers specialized in both the direction of a pedestrian and the distance of the pedestrian from a camera. Using this framework, the developed system can improve the detection performance and provide information of the direction of a pedestrian simultaneously. The experimental results show that superior detection performance and direction estimation results were obtained by our method.

Visibility Estimation under Night-time Conditions using a Multiband Camera

Various driver-assistance systems are currently being developed that make use of on-vehicle cameras. However, the imaging conditions and the methods used to detect objects are different for each system. Therefore, a special camera is often needed in order to satisfy the requirements of each system. A camera that can be shared by multiple systems will become essential when more systems are put to practical use in the future. Therefore, a multiband camera has been developed that can provide both color images and near-infrared images. The camera includes a special filter that improves on the Bayer filter arrays that are used in single-chip digital color cameras, and it can simultaneously obtain images covering four wavelength bands that have the same optical axes and fields of view. Moreover, a method for estimating the drivers visibility when using the camera is described in this paper.

Reliable pedestrian recognition combining high-definition LIDAR and vision data

Pedestrian recognition is one of the key technologies for advanced driver assistance systems and autonomous driving systems. The present paper proposes a fusion system for reliable pedestrian recognition using high-definition LIDAR and a vision sensor to achieve high performance under various conditions. Pedestrian candidates are extracted from two sensors in parallel by support vector machine-based classifiers. In particular, the region of interest in the image processing is set based on information about objects derived from the LIDAR processing in order to reduce false positives as well as the computational burden. All candidates are integrated by their likelihood, as calculated from their classification scores, using multiple thresholds according to the detection condition of the target in two sensors. A quantitative evaluation in a road environment confirms the effectiveness of the proposed system.

Multiband image segmentation and object recognition using texture filter banks

Current driving assitance systems have multiple cameras mounted on a moving vehicle for road environment perception. For the purpose of integrating a color camera and a near infrared camera, we developed multiband camera. An input image of the multiband camera, which is called a multiband image, is available in four bands consisting of a band of near infrared and three bands of color. In this paper, we present a multiband image segmentation to recognize the objects in a road scene using various texture filter banks. Experimental results show that the performance of a multiband image is superior to that of a color image for multiclass object recognition.

Improved Lane Detection Based on Past Vehicle Trajectories

Knowing where the host lane lies is paramount to the effectiveness of many advanced driver assistance systems (ADAS), such as lane keep assist (LKA) and adaptive cruise control (ACC). This paper presents an approach for improving lane detection based on the past trajectories of vehicles. Instead of expensive high-precision map, we use the vehicle trajectory information to provide additional lane-level spatial support of the traffic scene, and combine it with the visual evidence to improve each step of the lane detection procedure, thereby overcoming typical challenges of normal urban streets. Such an approach could serve as an Add-On to enhance the performance of existing lane detection systems in terms of both accuracy and robustness. Experimental results in various typical but challenging scenarios show the effectiveness of the proposed system.

Road ortho-image generation based on accurate vehicle trajectory estimation by GPS Doppler

This paper proposes a novel technique to generate accurate road ortho-images using low-cost sensors (single frequency GPS, speed sensor and MEMS yaw rate gyro). Ortho-images are data incorporated into accurate digital maps. Our proposal should contribute to reduction in production costs of accurate digital maps. The most striking feature of our proposal is vehicle trajectory estimation using GPS Doppler, for road image mosaicing. We realize accurate trajectory and heading angle estimation by integrating GPS Doppler and yaw-rate gyro. This enables creation of ortho-images with a combination of devices less expensive than the conventionally used ones. A comparison with commercial road ortho-images shows that our proposed technique can realize the same level of accuracy.

Learning-based trajectory generation for intelligent vehicles in urban environment

Recent technologies of intelligent vehicles are getting more attentions with promising deployment to commercial cars. In this paper, we present a learning-based trajectory generation approach for implementing an advanced driver assistance system (ADAS) with the lane keep assist and adaptive cruise control functions in urban environment. More specifically, a number of objects of interest, including the road and lane boundaries, as well as the surrounding vehicles, are detected and tracked. Particularly, the leader vehicle in the host lane, if available, is detected to provide the real-time, on-site and validated information, including both movements and decisions of how it copes with the current traffic situation, which is subsequently learnt by the ego vehicle for the control purposes. By combining the prior and “live” information of the road environment, a safe, smooth and reasonable trajectory is finally generated based on a cubic spline model with the Mass-Spring-Damper (MSD) system. Experimental results in various typical but challenging urban traffic scenes have substantiated the effectiveness of the proposed system.