Which is better or worse between LiDAR and pure computer vision for autonomous driving?

2024-01-04

Lidar and pure machine vision have always been two distinct directions in autonomous driving technology.

In practical applications, both have their own advantages and disadvantages in terms of data form, accuracy, cost, and so on.

The basic principles of autonomous driving technology

The auto drive system can be divided into the perception layer, the decision-making layer, and the executive layer.

The perception layer captures vehicle location information and external environmental information through various hardware sensors.

The "brain" of the decision-making layer models the environment based on the information input from the perception layer, forming a global understanding and making decision judgments, and then issuing signal instructions to the vehicle for execution.

The subsequent execution layer converts the signals from the decision-making layer into the action behavior of the car.

The LiDAR route and pure machine vision route are both ways for autonomous vehicles to perceive their environment, with the only difference being their implementation methods.

The machine vision route is dominated by cameras, combined with millimeter wave radar, ultrasonic radar, low-cost LiDAR, etc. The LiDAR route is dominated by LiDAR, combined with millimeter wave radar, ultrasonic sensors, cameras, etc.

The technical principles of LiDAR and cameras

Lidar, whose working principle is to use laser for detection and ranging technology, is usually installed on the top of cars and can monitor 360 degrees. Inside the LiDAR, each group of components includes a transmitting unit and a receiving unit.

The principle of distance measurement is similar to laser measurement of the distance between the Earth and the Moon, which is based on the time of laser emission and return. The laser diode emits pulsed light, which reflects a portion of the light back when it reaches the target object. A photon detector is installed near the diode to detect the returning signal, and the distance to the target object can be calculated by calculating the time difference.

After the pulse distance measurement system is activated, it can collect a large amount of point clouds, and if there is a target object in it, it will appear as a shadow in the point cloud. Point clouds can generate a 3D model of the surrounding environment. The higher the point cloud density, the clearer the image.

It can be considered that the two important attributes of LiDAR are ranging and accuracy. Unlike cameras, the LiDAR solution is "active vision" - it can actively detect the surrounding environment, and the strength of the ambient light is not important. It can work day and night. Meanwhile, thanks to the more concentrated laser beam, it has higher detection accuracy than millimeter wave radar.

The working principle of a camera is similar to that of a human eye, where the light reflected by an object is imaged on a sensor through a lens. Its disadvantage lies in its ranging ability and the significant impact of environmental lighting. At the same time, one of its great advantages is that people can intuitively understand the content captured by the camera, and using it to classify objects, that is, visual recognition work, will be very easy.

Singapore Stock Exchange and Temasek collaborate to streamline transaction processes using blockchai