This paper proposes a convolutional neural network (CNN)-based method that learns traffic as images and predicts large-scale, network-wide speed with high accuracy. Spatiotemporal dynamics are converted to describing the time space relations of flow via two-dimensional time-space matrix. A CNN is applied image following two consecutive steps: abstract feature extraction prediction. The effectiveness proposed evaluated by taking real-world transportation networks, second ring road north-east...

Path travel time estimation for buses is critical to public transit operation and passenger information system. State-of-the-art methods estimating path are usually focused on single vehicle with a limited number of road segments, thereby neglecting the interaction among multiple buses, boarding behavior, traffic flow. This study models considering link station dwell time. First, we fit shifted lognormal distributions as in previous studies. Then, propose probabilistic model capture...

Keypoint matching is an important operation in computer vision and its applications such as visual simultaneous localization mapping (SLAM) robotics. This heavily depends on the descriptors of keypoints, it must be performed reliably when images undergo condition changes those illumination viewpoint. Previous research keypoint description has pursued three classes descriptors: hand-crafted, from trained convolutional neural networks (CNN), pre-trained CNNs. paper provides a comparative study...

Abstract Cellular data is a sequence of base station‐interaction that records user ID, timestamp, location area code (LAC), and cell identity (CID). With long observation periods, the allows traffic planners to analyze coarse‐granularity travel behaviours at low costs. However, utilizing cellular for urban planning not an easy task as lacks socioeconomic attributes due privacy issues. The also challenging recognize activity types. This paper proposed activity‐based model (ABM) with skeleton...

The L2-normalization and feature pooling have a wide range of applications in image classification also achieved remarkable results. However, there is much room for the existing descriptors which are extracted from pre-trained Convolutional Neural Network (CNN) models, to meet requirement precision patch classification. We generate CNN by using on descriptors. By evaluation Brown dataset, mean Average Precision (mAP) descriptor, based Inception-v3 model that applies both pooling, reaches...

Motivated by the need to improve performance of visual loop closure verification via multi-view geometry (MVG) under significant illumination and viewpoint changes, we propose a keypoint matching method that uses landmarks as an intermediate image representation in order leverage power deep learning. In environments with various traditional MVG may encounter difficulty because their inability generate sufficient number correctly matched keypoints. Our exploits excellent invariance properties...

Understanding human travel patterns with relation to land uses and other characteristics is a crucial topic in urban studies offers guidance better design, plan manage diverse transportation systems infrastructures. However, previous on such still have some limitations. First, current mainly focus mobility pattern analysis stop route levels. Rare investigated user region levels exhibit macroscopic perspective activities. Second, the works usually examine one specific mode fail compare...

Abstract Existing studies on activity location recognition based mobile phone data has made great progresses. However, current generally assume constant distance threshold when performing clustering, and ignore the influence of base station layout positioning accuracies data. Given different accuracy requirements, authors propose two methods to recognise locations: (1) An improved hierarchical agglomerative clustering algorithm that integrates a genetic component search dynamically adjust...

Keypoint matching is an important operation in computer vision and its applications such as visual simultaneous localization mapping (SLAM) robotics. This heavily depends on the descriptors of keypoints, it must be performed reliably when images undergo conditional changes those illumination viewpoint. In this paper, a descriptor fusion model (DFM) proposed to create robust keypoint by fusing CNN-based using autoencoders. Our DFM architecture can adapted either trained or pre-trained CNN...

Large-scale traffic sensors are strategically deployed across various infrastructures and modes of transportation (e [...]

This paper proposes a convolutional neural network (CNN)-based method that learns traffic as images and predicts large-scale, network-wide speed with high accuracy. Spatiotemporal dynamics are converted to describing the time space relations of flow via two-dimensional time-space matrix. A CNN is applied image following two consecutive steps: abstract feature extraction prediction. The effectiveness proposed evaluated by taking real-world transportation networks, second ring road north-east...

For robust navigation, the objective in visual SLAM is to create a dense map from sparse input. Although there have been significant number of endeavors on real-time mapping, existing works for systems still fail preserve adequate geometry details that are important navigation. This paper estimates pixel-wise depth single image and few points which constructed registered LiDAR or acquired construct map. The main idea employ set new loss functions consisting photometric reconstruction...

Vehicle-based mobile sensing (a.k.a drive-by sensing) is an important means of surveying urban environment by leveraging the mobility public or private transport vehicles. Buses, for their extensive spatial coverage and reliable operations, have received much attention in sensing. Existing studies focused on assignment sensors to a set lines buses with no operational intervention, which typically formulated as covering subset selection problems. This paper aims boost power bus fleets through...