It’s common knowledge today that data labeling in general and in particular support numerous Machine Learning (ML) applications – from self-driving vehicles and e-commerce to machine translation and virtual assistants. There are, however, other domains where crowd contributors make a huge impact. Moreover, pretty much all of us use the results of their efforts every day, often without realizing it. crowdsourcing This series of articles will look at 3 common categories of spacial or “feet-on-street” crowdsourcing – along with the use cases – that improve our daily lives both as private citizens and consumers, as well as business owners. The first article will cover spatial crowdsourcing and web mapping; the second one will explain crowdsourcing in the context of brick-and-mortar retail; and the last article will dive into crowd-assisted strategies involved in verification of business information. Spatial crowdsourcing and field tasks for digital mapping Whether you’re looking to beat traffic by using a less-known travel route, deciding on a nearby diner that has decent reviews, searching for a place to get a haircut when you’re new in town, or trying to locate the nearest post office to drop off a parcel, you’re likely to turn to the same source – online maps. The field of digital mapping and the web mapping platforms that offer these services are everywhere these days, and much of what we see on these platforms is supported by crowdsourcing. Granted, there’s more than one way the data is put together in computer cartography in order to offer a working product to the end user. But when it comes to the all important last-mile outer layer of information – names of establishments, exact addresses, accessibility, position of on-site objects, etc — these are very often tackled by human labelers who bring us the latest and most accurate information. Although different crowd-assisted tasks may be applicable to mapping, arguably the most common category is “spatial crowdsourcing” aka “feet-on-street” or simply “field” task. Spatial crowdsourcing is performed offline, meaning that crowd contributors physically travel to target locations and carry out specific instructions. Below are some of the best known spatial crowdsourcing tasks and processes that support digital mapping by generating high-quality geo data. Numbering building entrances is a common feet-on-street process whose goal is to add crucial details to online maps, which is referred to as “map enriching.” As a result, both consumers and businesses alike can organize their activities and reliably visit and/or deliver products and services to a variety of destinations within a set area. To complete such field tasks, crowd contributors on the ground are provided with the area’s data, including any relevant plans and layouts, as well as the property’s address or GSC coordinates, which is known as the “input data.” The “output data” consists of enriched images with clearly marked entrances to the building. Updating and adding organizations is another common request by map developers and users. This is the main method of keeping information up-to-date, as it enables uninterrupted workflow of third-party couriers, taxi companies, logistics services, and, of course, the organizations themselves. To complete these tasks, contributors on the ground are given plans of the area along with any current business information available. After task completion, enriched maps will have updated information on existing companies and also fresh additions – either recently established businesses or previously unlisted entries. Importantly, up-to-date information always contains so-called “attributes,” which is specific information pertaining to organizations. These may include working hours, links to company websites and social media, reviews by clients, as well as other features such as payment methods, wheelchair access, price range, or type of cuisine in the case of food establishments. Gathering last-mile data refers to collecting information about various objects in the area, including but not limited to parking lots, stairs, fences, benches, and playgrounds. These are called “last mile” because these objects are the “finishing touches” information-wise, upping the end user’s satisfaction. Crowd contributors tackling such field tasks are provided with the area’s maps along with any available information about existing objects contained therein. After visiting these locations in person and marking the objects accordingly, urban environment data will include updated images depicting all relevant items within the target area, as well as their precise coordinates/positioning. Gathering radio signal data for shopping malls is in high demand because sometimes GPS signals are weak or there is interference. Regardless of this, real-time positioning of online map users has to always be available. So, to achieve this, signal information along set routes within shopping malls is collected. Field task contributors are provided with specific prearranged routes that they have to follow and measure signal strength. After the task is complete (which normally takes 3-4 hours), system information on signals at set frequencies is transmitted from the contributors’ devices back to those who process it. Creating shopping mall plans and updating info on layouts and occupants is especially relevant with newly established or renovated/refurbished shopping malls but also applies to all fully operational shopping malls that inevitably undergo some changes from time to time. The objective here is to have an updated floor plan that may include on-site outlets/brands, bathrooms, stairs, water fountains, food courts, playgrounds, and so on. To deliver this information, feet-on-street contributors are given coordinates of shopping malls along with any existing floor plans if such are available. After 3-7 days (depending on the mall’s size and availability of information prior to the task), updated floor plans are usually ready to be assimilated into mapping services. Collection of pedestrian panoramas is a practical way of obtaining useful geo details, as well as getting a more personal feel for public areas, such as parks, squares, heritage sites, and business districts. Today, 360° panoramas are gradually replacing older, lower-quality passing car or aerial image methodologies for close-up imagery. With these tasks, crowd contributors on the ground are given coordinates and asked to visit certain sites and take panoramic photos, usually with their phones. Each site requires around 400 data points to form a high-quality, cursor-draggable panoramic image for the end user. These data points normally come from multiple contributors and are put together via . aggregation Collecting information about terminals and delivery points is crucial for speedy and effective functioning of courier services, as well as time management of private citizens. Counterintuitively, parcel pick-up/drop-off points are sometimes found in hard-to-reach and obscure spots, which is why laying out clear routes to these locations can help avoid unnecessary meanderings and confusion. Field task contributors in this case are provided with whatever directions are available, at which point they proceed to the site, take photos, and work out the best routes to the location. Subsequently, digital maps are updated with arrowed visual directions and/or descriptions of “best access” in the form of text. Collection of automobile panoramas is similar to obtaining pedestrian panoramas, except that it’s done from the contributors’ vehicles that follow preset travel routes. Using the route’s coordinates, crowd contributors take and tag photos from their two phones (a prerequisite for the task) as they drive through multiple locations. This method works especially well in towns and cities with a population of up to 100+K in lieu of the standard street-view data collection vehicles. As a result, two important goals are achieved: (1) high-quality panoramic images of different travel routes and (2) more information about organizations in the area, including their names, addresses, and business categories (which supplements and enhances “updating and adding organizations” mentioned earlier that’s carried out on foot). Obtaining weather data through crowdsourcing serves a practical purpose, because forecasts don’t always offer a realistic estimate. Trusted feet-on-street residents are called upon to observe, assess, and describe weather in real time around their home regions. It works in the following way: crowd contributors (permanent or temporary inhabitants of city boroughs, districts, or neighborhoods) are asked to look at their physical thermometers to determine local temperature and also report on how it feels to them (i.e., “real feel”®). This is accompanied by additional multiple-choice questions that address specific climate features such as precipitation, wind, visibility, etc. Afterthought As we’ve seen, crowd-based methodologies work well to complement existing online mapping solutions, and they’re effective in data collection (particularly last-mile), data cleansing/comparison, and ultimately digital map enrichment. In our next article of the series, we’ll look into crowdsourcing from the perspective of brick-and-mortar retail –  how crowd contributors can help make information more accurate and transparent for the end user, be it a customer or a proprietor.

Spacial Crowdsourcing and Its Applications: Web Mapping

Too Long; Didn't Read

People Mentioned

@fedulov

RELATED STORIES