Dockless electric scooters (e-scooter) have emerged as a green alternative to automobiles and a solution to the first- and last-mile problems. Demand anticipation, or being able to accurately predict spatiotemporal demand of e-scooter usage, is one supply–demand balancing strategy. In this paper, we present a dockless e-scooter demand prediction model based on a fully convolutional network (FCN) coupled with a masking process and a weighted loss function, namely, masked FCN (or MFCN). The MFCN model handles the sparse e-scooter usage data with its masking process and weighted loss function. The model is trained with highly correlated features through our feature selection process. Next-hour and next 24-h prediction schemes have been tested for both pick-up and drop-off demands. Overall, the proposed MFCN outperforms other baseline models including a naïve forecasting, linear regression, and convolutional long short-term memory networks with mean absolute errors of 0.0434 and 0.0464 for the next-hour pick-up and drop-off demand prediction, respectively, and the errors of 0.0491 and 0.0501 for the next 24-h pick-up and drop-off demand prediction, respectively. The developed MFCN expands the collection of deep learning techniques that can be applied in the transportation domain, especially spatiotemporal demand prediction.

This paper presents a new approach for text-based video content retrieval system. The proposed scheme consists of three main processes that are key frame extraction, text localization and keyword matching. For the key-frame extraction, we proposed a Maximally Stable Extremal Region (MSER) based feature which is oriented to segment shots of the video with different text contents. In text localization process, in order to form the text lines, the MSERs in each key frame are clustered based on their similarity in position, size, color, and stroke width. Then, Tesseract OCR engine is used for recognizing the text regions. In this work, to improve the recognition results, we input four images obtained from different pre-processing methods to Tesseract engine. Finally, the target keyword for querying is matched with OCR results based on an approximate string search scheme. The experiment shows that, by using the MSER feature, the videos can be segmented by using efficient number of shots and provide the better precision and recall in comparison with a sum of absolute difference and edge based method.

In this paper, a region-based text localization that is robust for multiple languages is presented. Maximally Stable Extremal Regions (MSERs) are used for detecting candidates of text areas. The MSER components are grouped based on their connectivity in a feature space by using a new proposed rule for assigning the connectivity. The groups of components are classified into three classes that are text regions with high confidence, text region with low confidence, and non-text regions. A chain of text attribute constraint decision with the double-threshold scheme is developed to identify text regions. A sequence of constraint decision is designed to minimize the complexity based on short-circuit evaluation of logic operators. The regions that satisfy all strong constraints will be considered as text regions with high confidence while the regions that fail in some strong constraints but satisfy all weak constraints will be considered as text regions with low confidence. The final text regions are obtained from all text regions with high confidence and text regions with low confidence that have connectivity to text regions with high confidence. The proposed scheme is evaluated by using the natural scene images that consist of totally nine languages with different text alignments and camera views. The experiment shows that our proposed scheme can provide the satisfy results in comparison with baseline method.

Marker registration is a fundamental process to estimate camera poses in marker-based Augmented Reality (AR) systems. We developed AR system that creates correspondence virtual objects on handwritten text markers. This paper presents a new method for registration that is robust for low-content text markers, variation of camera poses, and variation of handwritten styles. The proposed method uses Maximally Stable Extremal Regions (MSER) and polygon simplification for a feature point extraction. The experiment shows that we need to extract only five feature points per image which can provide the best registration results. An exhaustive search is used to find the best matching pattern of the feature points in two images. We also compared performance of the proposed method to some existing registration methods and found that the proposed method can provide better accuracy and time efficiency.