| MAIPRADIT ARNAN | D, 中間発表 | ユビキタスコンピューティングシステム | 安本 慶一, | 池田 和司, | 諏訪 博彦 |
|
title: Self-Learning Method for Vehicle Detection with Vibration Sensor and Video Camera based on Q-Learning.
abstract: Vehicle detection systems play a significant role in ITS (Intelligence Traffic System), due to their wide applicability to traffic analysis, vehicle counting and vehicle classification. However various variables like weather condition, speed of vehicle, shape, shadow, a blur of camera and obstacle, cause inaccuracy in the detection system which leads to ineffectiveness of the traffic management We propose a self-learning vehicle detection system with a vibration sensor and video camera based on Q-Learning. The system automatically collects the ground truth labels from video camera data using image analysis . We collect data for detection by using a vibration sensor in the same way as traffic census sensor. Additionally, Q-learning, one of the model-free reinforcement learning, is used for real-time self learning. With the proposed system, the passing vehicle detection model using vibration data is automatically trained by setting the camera and the vibration sensor on the roadside. The trained model is used to detect passing vehicles without cameras once it is trained. language of the presentation: English | |||||