Basics of LiDAR imaging

Regarding the technology of detecting the distance of an object, we have always heard of RADAR. This device can calculate the position of objects in space by relying on the bounce of radio waves. Thanks to the growth of electromagnetic wave technology, only in recent years, with the concept of RADAR technology plus some optical principles, researchers have developed a technology that can more accurately measure distance, that is, LiDAR. In other words, the RADAR system emits radio waves and measures the bounce signal. LiDAR uses light waves when measuring distance. Because the data collection capabilities of this system are more powerful, it can provide more practical 3-D information models. Li3-DAR has a wide range of uses and is suitable for different purposes. Especially in the last five to ten years, many companies and academic circles have begun to focus on this technology[12]. More precisely, LiDAR is a distance

measurement method that measures distance through a measuring beam. LiDAR uses pulsed laser light to measure objects and uses extremely accurate detectors to collect the returned signal light. The calculated data can then be used to digitize the surrounding environment to obtain a 3-D representation.

As shown in Fig. 1.3, the beam divergence angle depends on the ratio of the aperture and wavelength of the transmitting antenna (radar) or lens (LiDAR). This ratio is too large for the larger beam divergence angle and smaller angular resolution produced by radar.

As shown in the figure above, radar (black dashed line) cannot distinguish between the two vehicles, while LiDAR (red dashed line) can. Radar's angular resolution is low, and he may regard the two closer cars as one large car, thus sending out wrong signals and making identification difficult[13]. The recognition ability of LiDAR is very high, and pedestrian gestures from tens of meters away can be easily recognized.

Fig. 1.3 Schematic diagram of LiDAR and RADAR application in autonomous driving. Radar's angular resolution is low, and it may treat two closer cars as one large car, thus sending out wrong signals and making identification difficult. The recognition

ability of LiDAR is very high, pedestrian gestures from tens meters away can be easily recognized.

Fig. 1.4 Depth image example. Through the depth image, we can know the distance between each object and the scanner directly from the image.[14]

LiDAR mostly uses infrared light to image objects. It can detect many target objects, including cars, rocks, non-metallic materials, rain, clouds, aerosol, and even single chemical molecules[12, 15, 16]. But we must keep one thing in mind, because LiDAR uses the reflected light signal from the surface of the object to image, so we can only

understand the surface of the target, but cannot penetrate the object. Therefore, it cannot penetrate dense smoke. In any case, the laser beam can present the uneven surface of the object with a high resolution. The beam reflection signal collected by LiDAR is not pure reflection, but there are many types of scattering, the most common is Rayleigh scattering[17].

Through LiDAR, we can get the distance between the light source and the target object. At this time, the obtained detection data is in one-dimensional form, that is, many different distance data will be received. Through the processing of the image processing program, the obtained distance data can be converted into a three-dimensional image. The denser the data points can form a more accurate model, or use the SP-line mathematical calculation method to reconstruct the surface shape of the object[10, 12]. This process is called 3D reconstruction. Unlike a camera, the image formed by LiDAR has depth data, also called depth image[18, 19]. The camera only captures the color information in the space. With the technology that can detect the distance of objects, this technology can be used in many places. For example, autonomous driving, robotics, factory automation, etc.

Fig. 1.5 Application of LiDAR in ADAS system.[2]

Three-dimensional measurement is a traditional technique for measuring three-dimensional images in space. When the LiDAR technology was not yet mature, two cameras were used to shoot left and right, and a three-dimensional model was constructed using visual stereo imaging. After filmmakers used high-definition cameras to make 3D movies many years ago, many 3D imaging products entered the consumer market. In recent years, due to the improvement of mobile phone performance, many game manufacturers have launched games with 3-D display functions, which greatly increases the demand for 3-D measurement technology. The other three requirements for measurement are industrial manufacturing industries, such as industrial inspection, reverse engineering, robots, and automatic vehicles[20]. In the process of monitoring and automation that requires higher precision, traditional measurement methods can no longer meet today's needs. Imaging measurement size and target recognition require better hardware architecture to achieve higher resolution and computing speed. When moving like a mobile robot, it requires three-dimensional measurement of the environment map, real-time dynamic scanning of the scene, simultaneous localization and mapping (SLAM)[21]. When the automatic driving system or robot moves, it uses the geometric features (edges, angles) of environmental objects to locate its own position, so that it can grasp the relative position of adjacent objects, such as walls, pillars, road trees, etc., and move instantly. If there are only two-dimensional images, it is difficult to correctly perceive the real spatial relationship between the environment and the object, and the system cannot make correct decisions and actions. Therefore, in addition to traditional 3D film and television products or reverse reconstruction engineering applications, 3D measurement has become a key technology in industries such as automation, robotics, and self-driving cars.

In ToF LiDAR measurement, the laser light source can only measure the distance of a single point from emission to reception. If you want to measure the data of the entire spatial range, you need to deflect the beam through some mechanical structure or optical technology. There are many different methods to deflect the laser light to different positions to scan the complete field of view. The method of deflecting laser light will be explained in detail in chapter 2.2. But when reconstructing a 3D image, you need to know the relative position of each data point in space. By combining the data of each signal with the scanning position, that is, combining the distance and the data of the x-axis, y-axis and z-y-axis in the space, all the data points can be integrated into a point cloud diagram that is convenient for observation.

Table 1.1 Comparison of three 3D imaging methods.