Developing an Interactive Machine-Learning-Based Approach for Sidewalk Digitization

016679572018PDFTech ReportNCST-RR-2018-1Pedestrian trafficBicycle travelBicycle commutingAdvanced traveler information systemsAdvanced traffic management systemsITS program technologiesSidewalksBikewaysDigital mappingImage processingLearning (Artificial intelligence)Advanced pedestrian information systemsAdvanced bicyclist information systemsMachine-learningResearch HubData and Information TechnologyHighwaysPedestrians and BicyclistsPlanning and ForecastingCaliforniaUnited StatesNational Center for Sustainable TransportationLuo, JiWu, GuoyuanUniversity of California, Riverside. Center for Environmental Research and TechnologyNational Center for Sustainable TransportationUS Transportation CollectionIn urban areas, many socio-economic concerns have been raised regarding fatal collisions, traffic congestion, and deteriorated air quality due to increased travel and logistic demands as well as the existing on-road transportation systems. As one of the promising remedies, active transportation has been advocated, which may not only mitigate congestion on local streets, but also promote physical fitness, foster community livability, and boost local economy (i). To promote the active transportation mode, extensive work has been focused on planning and developing a number of pedestrian and bicyclist related programs which require the infrastructure, e.g., sidewalks, as a premise (ii). A significant amount of these efforts have to go for the setup, maintenance and evaluation of the sidewalk inventory on a relatively large geographic scale (e.g., citywide, statewide), which lays a solid foundation for a variety of active mobility-focused applications and related research, for example: Improved Location-awareness Service. As illustrated in Figure 1, the state-of-the-art navigation tools (e.g., Google Maps [https://www.google.com/maps]) rely on the roadway network which is designed for vehicles to guide pedestrians. Some navigation instructions can be confusing and even pose safety risks to those vulnerable road users, since a portion of the path may not be walkable or be in conflict with motor vehicles. In such cases, a dedicated sidewalk network becomes necessary. Crowdsourcing Based Sidewalk Inventory Maintenance and Update. As aforementioned, a large-scale sidewalk inventory will facilitate the planning of new sidewalk construction, and maintenance or improvement of existing sidewalks (iii). For example, based on the sidewalk network, active travelers and traffic engineers can identify or report damaged sidewalks and share the locations of potential safety risks in a timely and cost-effective manner. Conventionally, transportation engineers and researchers have to rely on laborious field measurements to conduct sidewalk survey and assessment (iv, v), which is rather resource consuming (in both time and cost). Recently, a few studies attempted to digitize sidewalks as a part of geographic information system (GIS), created the sidewalk inventory under restricted conditions (vi), and assessed the quality of sidewalk (vii, viii). However, most of the existing methods for sidewalk system digitization are neither comprehensive nor cost-effective. On the other hand, due to the rapid advances in computational capability and explosion of data availability, machine learning techniques have shown great potential for image recognition and classification (ix). One heuristic (brute-force) way to extract the information of sidewalks is to process every piece of satellite/aerial images (with appropriate size and resolution) for the region of interest. However, it may be overwhelming to prepare the image set (for both training and analysis), especially for a large geographic scale. In addition, it would be also very challenging to develop a reliable and effective algorithm to identify the sidewalk from a large mix of images on different facilities. To address the above issues, we propose a machine-learning-based (x, xi) sidewalk digitization method which should be much more reliable and cost-effective than the brute-force one. The basic idea is to take full advantage of roadway networks to reconstruct an initialized (connected) sidewalk network. Then, an image sweeping script is developed to extract a large number of sidewalk images along the initialized sidewalk network. Thirdly, a machine learning technique is applied to the aerial images of focused areas (i.e., surrounding zones along the initial s) Conventionally, transportation engineers and researchers have to rely on laborious field measurements to conduct sidewalk survey and assessment (iv, v), which is rather resource consuming (in both time and cost). Recently, a few studies attempted to digitize sidewalks as a part of geographic information system (GIS), created the sidewalk inventory under restricted conditions (vi), and assessed the quality of sidewalk (vii, viii). However, most of the existing methods for sidewalk system digitization are neither comprehensive nor cost-effective. On the other hand, due to the rapid advances in computational capability and explosion of data availability, machine learning techniques have shown great potential for image recognition and classification (ix). One heuristic (brute-force) way to extract the information of sidewalks is to process every piece of satellite/aerial images (with appropriate size and resolution) for the region of interest. However, it may be overwhelming to prepare the image set (for both training and analysis), especially for a large geographic scale. In addition, it would be also very challenging to develop a reliable and effective algorithm to identify the sidewalk from a large mix of images on different facilities. To address the above issues, we propose a machine-learning-based (x, xi) sidewalk digitization method which should be much more reliable and cost-effective than the brute-force one. The basic idea is to take full advantage of roadway networks to reconstruct an initialized (connected) sidewalk network. Then, an image sweeping script is developed to extract a large number of sidewalk images along the initialized sidewalk network. Thirdly, a machine learning technique is applied to the aerial images of focused areas (i.e., surrounding zones along the initial sidewalk network) to identify whether a sidewalk is present or not. It is noted that the category of sidewalk (e.g., landscape/lawn, parking lot/ driveway, crosswalk) may be also recognized if there were sufficiently large training dataset. The rest of this report is organized as follows: Section 2 will give an overview of the proposed methodology for sidewalk digitization. Section 3 and 4 will introduce the mapping and categorizing algorithms in details. Section 5 will present the application of the method on streets in Riverside City, and evaluate the performance of the method. The last section concludes this report with a discussion on potential future work