Page 17 - 卫星导航2021年第1-2合期

El‑Sheimy and Li Satell Navig (2021) 2:7 Page 7 of 23

unveils a new iPad Pro with a LiDAR scanner, which may (Wolcott and Eustice 2014) and (McManus et al. 2013)
bring new directions to indoor PLAN. respectively use images from monocular and stereo cam-
LiDAR measurements are used for PLAN through 2D eras to match the 3D point cloud map generated by a sur-
or 3D matching. For example, the research works (de vey vehicle equipped with 3D LiDAR scanners.
Paula Veronese et al. 2016) and (Wolcott and Eustice
2017) match LiDAR measurements with a 2D grid map RADAR
and a 3D point cloud map, respectively. Te PLAN per- RADAR has also received intensive attention in the
formance is generally better when the surrounding envi- autonomous driving industry. Similar to LiDAR, the
ronment features are signifcant and distinct from other RADAR determines the distance by measuring the
places; otherwise, performance is limited. Te LiDAR round-trip time diference of the signal. Te diference
measurement performance will not be afected by light is that the RADAR emits radio waves, instead of laser
but may be afected by weather conditions. waves. Compared with LiDAR, the RADAR generally has
a further measurement range. For example, the Bosch
Camera LRR RADA can reach up to 250 m. Also, the price of a
Cameras are used for PLAN and perception by collect- RADAR system has dropped to the order of $ 1,000 to
ing and analyzing images. Compared with LiDAR and $ 100. Moreover, RADAR systems are lightweight, which
RADAR, the camera has a much lower cost. Also, the makes it possible to embed them in cars.
camera has the advantages such as rich feature informa- On the other hand, the density of RADAR measure-
tion and color information. Also, the camera is a passive ments is much lower than that of LiDARs and cameras.
sensing technology, which does not transmit signals and Terefore, RADAR is often used for obstacle avoid-
thus does not have errors on the signal-propagation side. ance, rather than as the main sensor of PLAN. Similar
Moreover, the current 2D computer vision algorithm is to LiDAR, the measurement performance of RADAR
more advanced, which has also promoted the application is not afected by light but may be afected by weather
of cameras. conditions.
Similar to LiDAR, the camera depends on the sig-
nifcance of environmental features. Also, the camera is WiFi/BLE
more susceptible to weather and illumination conditions. WiFi and BLE are the most widely used indoor wireless
Its performance degrades under harsher conditions, such PLAN technologies for consumer electronics. Te com-
as in darkness, rain, fog, and snow. Tus, it is important monly used observation is RSS (Zhuang et al. 2016), and
to develop camera sensors with self-cleaning, longer the typical positioning accuracy is at meter-level. Also,
dynamic range, better low light sensitivity, and higher researchers have extracted high-accuracy measurements,
near-infrared sensitivity. Furthermore, the amount of raw such as CSI (Halperin et al. 2011), RTT (Ciurana et al.
camera data is large. Multiple cameras on an autonomous 2007), and AoA (Quuppa 2020). Such measurements
vehicle can generate gigabyte-level raw data every minute can be used for decimeter-level or even centimeter-level
or even every second. PLAN.
Some PLAN solutions use cameras, instead of a high- A major advantage of WiFi systems is that they can
end LiDAR, to reduce hardware cost. An example is use existing communication facilities. In contrast, BLE
Tesla’s autopilot system (Tesla 2020). Tis system con- is fexible and convenient to deploy. To meet the future
tains many cameras, including three forward cameras Internet-of-Tings (IoT) and precise localization require-
(wide, main, and narrow), four side cameras (forward ments, new features have been added to both the latest
and rearward), and a rear camera. To assure the PLAN WiFi and BLE technologies. Table 6 lists the new WiFi,
performance in the environments that are challenging for BLE, 5G, and LPWAN features that can enhance PLAN.
cameras, RADARs and ultrasonic sensors are used. WiFi HaLow (WiFi-Alliance 2020) and Bluetooth long
Te two main camera-based PLAN approaches are range (Bluetooth 5) (Bluetooth 2017) are released to
visual odometry/SLAM and image matching. For the improve the signal range, while WiFi RTT (IEEE 802.11
former, the research work (Mur-Artal and Tardós 2017) mc) (IEEE 2020) and Bluetooth direction fnding (Blue-
can support visual SLAM using monocular, stereo, and tooth 5.1) (Bluetooth 2019) have been released for preci-
Red–Green–Blue-Depth (RGB-D) cameras. For image sion positioning.
matching, road markers, signs, poles, and artifcial fea-
tures (e.g., Quick Response (QR) codes) can be used. Te 5G/LPWAN
research work (Gruyer et al. 2016) uses two cameras to 5G has attracted intensive attention due to its high speed,
take the ground road marker and match it with a preci- high reliability, and low latency in communication. Com-
sion road marker map. In contrast, the research works pared with previous cellular technologies, 5G has defned