Shintaro Watanabe

In-vehicle camera traffic sign detection and recognition

In this paper, we discuss theoretical foundations and a practical realization of a real-time traffic sign detection, tracking and recognition system operating on board of a vehicle. In the proposed framework, a generic detector refinement procedure based on mean shift clustering is introduced. This technique is shown to improve the detection accuracy and reduce the number of false positives for a broad class of object detectors for which a soft response’s confidence can be sensibly estimated. The track of an already established candidate is maintained over time using an instance-specific tracking function that encodes the relationship between a unique feature representation of the target object and the affine distortions it is subject to. We show that this function can be learned on-the-fly via regression from random transformations applied to the image of the object in known pose. Secondly, we demonstrate its capability of reconstructing the full-face view of a sign from substantial view angles. In the recognition stage, a concept of class similarity measure learned from image pairs is discussed and its realization using SimBoost, a novel version of AdaBoost algorithm, is analyzed. Suitability of the proposed method for solving multi-class traffic sign classification problems is shown experimentally for different feature representations of an image. Overall performance of our system is evaluated based on a prototype C++ implementation. Illustrative output generated by this demo application is provided as a supplementary material attached to this paper.

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.