Outdoors, satellite positioning can be used for reasonably accurate and reliable positioning in open areas: even consumer devices can typically reach an accuracy of around five metres. The situation changes in urban areas and indoors, as tall buildings cause the satellite signals to reflect and attenuate. This will either prevent the tracking of satellite positioning signals or make it highly unreliable. Alternative technology is required to generate spatial data for these environments.
Some navigation technologies are absolute, reporting the location of the user in relation to a known coordinate frame or map. Other technologies are relative, and they only measure movement in relation to the previous location. The ideal solution is to seamlessly combine different technologies so that smartphone users will see the same positioning performance in different environments, for example. No attempts at this have been fully successful.
Most of us have come across positioning applications outdoors, the most common example probably being in-car navigation. Many of us have also probably shared our location through social media, e.g. at a café. So far, there are few, if any, applications for navigating indoors with the same accuracy as in-car navigation. It is not difficult to imagine different cases for its use: customers in a shopping centre want to find the way to a certain shop, perhaps even the correct shelf, and the vehicles of customers arriving by autonomous car will need to find their way around the car park.
In some cases, the data generated by indoor positioning may be of interest to someone other than the located subject. The safety of guards, nurses, and firefighters can be improved by automatically sent support in case they come under threat at some point in their duties. On the other hand, knowing the most popular shops and shelves is valuable information for the owners of shopping centres and shops.
The weak transmission power of satellite positioning signals makes them unsuitable for use indoors or between tall buildings. However, these locations often have other radio transmitters with more powerful signals that can be used to infer spatial data, even though that is not their original purpose. The term for these signals is signals of opportunity. Unlike motion sensors, radio signals can usually provide spatial data as absolute coordinates instead of in relation to a previous location.
Many homes and other buildings have wireless local area network (WLAN, "Wi-Fi") base stations, which can typically remain stationary for years. The WLAN signal always carries the base station's unique identifier, so a base station with a known location can act as a reference for estimating the location of the user. In many cases, WLAN-based positioning uses a radio fingerprint map, which is a database of the WLAN base stations and their signal strengths [KM1] available in different locations. This eliminates the need to know the locations of the individual base stations. Smartphones can listen to the WLAN signals around them and refer to the fingerprint database to infer their location. Bluetooth signals can also be used the same way. Today, crowdsourcing is used to compile and maintain these databases: different devices automatically send data to the manufacturer about their location and the signals they have detected, updating the database.
The same principle can be applied on a larger scale than local networks by harnessing mobile communication networks for positioning – their base stations can also be identified by their signals. The wider spread of mobile network base stations makes for worse accuracy, at best dozens or hundreds of meters, or even less in rural areas. This is expected to change over time as 5G networks come online, because 5G signals make direct ranging possible.
Other types of radio frequency-based ranging can be used, in professional applications in particular, including ultra wideband (UWB) methods that are designed for optimal wall penetration. In addition to radio frequencies, other signals such as ultrasound can be used to transmit absolute spatial data indoors.
Motion sensors allow the movement of the user to be measured regardless of their environment. As their name suggests, they only measure movement, and the origin coordinates must be determined otherwise.
Submarines and aeroplanes have traditionally relied on inertial navigation in situations where satellite positioning is unavailable. Changes in the user's location and direction of movement can be determined using accelerometers and gyroscopes. Today, MEMS (microelectromechanical system) technology allows accelerometers and gyroscopes to be manufactured cost-effectively and in tiny sizes; in practice, all smartphones include these sensors to determine the orientation of the device (to set the displayed image horizontally or vertically) and for gaming use. However, the error in position of even the highly expensive inertial sensors used in aeroplanes can grow to more than a kilometre over one hour of inertial navigation, if no other navigation methods are used.
If the location and velocity of the user are known at a certain point in time, inertial measurements can be used to determine their position and velocity at other times, at least in principle. The problem is the accumulation of errors in the measurements: for example, even a small bias error in velocity measurements results in an error in position relative to the time squared. The values reported by micromechanical sensors are especially prone to errors, and they are basically useless for conventional inertial navigation.
Motion sensors can be utilised in other ways than just conventional inertial navigation. For instance, the walking rhythm of a pedestrian can be identified from the impacts reported by accelerometers. If their direction of movement is also known in this case, the location of the user can be calculated based on the assumed length of one step. Here the error in position is relative to the distance travelled and not time to the second power, resulting in the error accumulating much more slowly. The direction of movement can be measured with gyroscopes or even compasses, i.e. magnetometers. Compasses are also typically unreliable indoors, as electronic equipment and ferromagnetic objects, such as iron reinforcement in concrete, will distort the magnetic field and interfere with the compass. On the other hand, these distortions of the magnetic field are often long-lived, allowing them to be mapped for positioning similarly to WLAN signal fingerprints.
The vertical coordinate is often more important indoors than it is outdoors, as in many cases information about indoor locations is useless if it points to the wrong floor. Barometers can be used to determine height, because the ambient pressure decreases as height increases. The problem with barometric measurements is that they are also affected by many environmental factors, such as weather and the air conditioning of the building. This means barometers require frequent calibration to provide reliable estimates of height.
Eyes are the most important navigation tool for people, so it follows that computer vision is useful for positioning. Images from a camera can be used in two ways: to identify landmarks and to track the movement of the camera. In the first method, the image is compared to a database of known views and the user's location can be determined if a match is found. These databases are labour-intensive to build and maintain, not to mention their storage space requirements, but data from services like Google Street View can make this positioning method viable on roads and streets.
The method based on tracking the movement of the camera requires no database but cannot report an absolute position and its coordinates. This type of motion tracking is based on comparing sequential images: the change in the position and orientation of the camera can be determined based on the change in the locations of the features recognised in the images. The accuracy of this method is diminished if these features are in motion, e.g. people or vehicles. Another problem is that a single camera cannot be used to measure distances directly, as the scale is not known. One picture is not enough to determine if an object is large and far away or small and near. The problem of scale can be solved with the use of a stereo camera, but this increases the physical size of the positioning equipment, as the lenses of the camera must be sufficiently far apart for accurate range imaging. Another solution is to use other motion sensors or prior information about the scale of the image (e.g. the distance of the camera from the ground or the dimensions of an object).
Active imaging devices, including lidar, are another way to implement computer vision. These methods have the advantages of also working in the dark and providing accurate ranging. They provide a three-dimensional image of the surrounding environment, even allowing mapping in addition to positioning (SLAM, simultaneous localisation and mapping). Lidar is not as widespread as cameras, motion sensors and radios, so they are not a good positioning solution for smartphone users, for example. In military applications, the beam shot out by the laser can also reveal the user to the enemy.
Indoor positioning needs multiple sensors
Typically, no single measuring technique or positioning system is sufficient for indoor positioning use in various operating environments. When different positioning methods are combined, it is vital to know the properties of the error sources in different measurements – for example, this allows the progressively growing inertial navigation error of velocity measurements to be distinguished from the quick shifts in estimates resulting from errors in a WLAN radio fingerprint map. Statistical filtering is a common approach where position is calculated based on probability distributions.
Prior information is also often used to support positioning when different measurements are combined. For example, the velocity of customers in a shopping centre is unlikely to exceed 10 km/h, even during Christmas sales, while the players in a basketball game tend to move fast. Map data is also useful, as it can show the areas users can access in the first place: if the estimate puts the user inside a wall, their position can be corrected to be next to the wall.