The core of autonomous vehicles is ADAS systems such as cameras and lidar.
Domestic company developing ADAS-based vision chip to be used in autonomous vehicles

If there is a technology that best fits the phrase, “dreams become reality,” it would be the self-driving car. When you call a car, it comes to you on its own, and it drives to its destination on its own. Self-driving cars recognize and avoid obstacles while driving, and provide various conveniences and information to the owner. Although it will take some time for fully self-driving cars to be commercialized, the so-called “semi-self-driving car” technology is becoming increasingly sophisticated. Now, it is easy to see advertisements for cars with self-driving functions, and they have become a deep part of our lives. This magazine will serialize the self-driving car project over the next 11 issues. Starting with the self-driving car industry trend, it will cover semiconductor components such as radar, lidar, and camera sensors, communications, precision maps, SW platforms, artificial intelligence, security, and K-City services. We ask for your continued support. <Editor’s Note>

As the commercialization of Level 3 autonomous vehicle technology begins, interest in advanced driving assistance systems is also increasing. ADAS systems are the foundation for autonomous vehicles, and in the beginning, they mainly played a role in informing road conditions through notifications, but recently, they have reached the level of automatically controlling the steering wheel, brakes, etc.

The Genesis EQ900 is equipped with advanced driving assistance systems (ADAS), including Advanced Smart Cruise Control (ASCC), Blind Spot Detection (BSD), Autonomous Emergency Braking (AEB), Driver Attention Alert (DAA), Smart High Beam Assist (SHBA), and Front Seat Pre-Active Seat Belts. The new i30 wagon is equipped with ADAS technologies such as adaptive cruise control (ACC), AEB, and lane keeping assist system. For such ADAS to operate, detection through cameras, lidar, and radar sensors is important.

Looking at the ADAS system installation by commercial vehicle, except for some small vehicles, cameras alone or radar and stereo cameras are applied. The range that the camera can cover is 100m in front, surround view, and rear. It plays a role in intuitively helping the driver because it can be seen with the actual eyes. When autonomous driving is achieved, at least 5 camera sensors per vehicle must be installed to detect 360-degree full coverage.

Vehicle cameras use 1 megapixels. 2 megapixel products are under development and prototypes have been released. Objects can be detected up to 100m away, and pedestrians up to 40m away. Depending on the type, they are divided into daytime and nighttime types, and are divided into single and stereo types. Stereo types use two cameras and can obtain distance information.

WDR(Wide Dynamic Range) CIS sensor provides characteristics similar to human vision. The purpose is to detect objects stably in situations such as backlighting or passing through tunnels while driving. NIR night vision sensor is to prevent accidents caused by collisions with wild animals at night, and can distinguish objects based on temperature differences when taking images. Stereo vision is trending at 16cm between the two cameras, but large commercial vehicles use ones with wider spacing, and general passenger cars use ones with narrow spacing. It is used to check how far the road is and whether you are leaving the road rather than directly detecting the distance. It expresses the actual distance information as a depth image.

The leader in automotive camera sensors is Israel's Mobileye. It currently supplies software and chips to more than 80% of OEMs worldwide, and is researching and developing self-driving cars with most companies. Mobileye has already reached agreements with some companies to supply image devices applicable to level 4 autonomous driving, and EyeQ5 and REM are scheduled to be released in 2018.

In Korea, Nextchip is developing APACHE4, a vision-based ADAS SoC. Apache is an integrated chip that can be applied to a vision-based ADAS system with an ADAS algorithm added, built-in ETRI's vehicle CPU, 'Aldebaran'. It is capable of pedestrian detection (PD), vehicle detection (VD), lane detection (LD), and moving object detection (MOD).
Nextchip’s Jeong Hoe-in, head of the research lab, said, “We are currently testing Apache 4. We are checking and resolving issues that have not yet been resolved through road tests,” and “We are also considering applying deep learning to some extent in the next lineup.” Apache 4 is scheduled to be prototyped early next year and enter mass production in 2020, and will be supplied through Tier 1 vendors of automakers.


Lidar, trend toward miniaturization and low-cost development

Another essential sensor for securing 3D location information in autonomous vehicles is LIDAR. The global LIDAR market was valued at $1.4279 billion in 2016. It is expected to grow at a compound annual growth rate (CAGR) of 25.8% to reach $5.248 billion in 2022.

Lidar technology can be implemented in various ways, and is divided into active technology that irradiates lasers and passive technology that does not irradiate lasers. To be suitable as a vehicle Lidar, it must be usable both during the day and at night, and be capable of detecting high-resolution 3D spatial information about the surroundings at a range of 100 m or more.

The lidar model initially applied to Google's self-driving car is Velodyne's HDL-64, a rotating scanning lidar sensor with 64 laser channels. It has the disadvantage of being difficult to commercialize, with a price range of 100 million won. Accordingly, Google miniaturized the lidar sensor while making its own self-driving car in 2014.

Ford demonstrated a self-driving car capable of night driving in 2016. This was to show that night driving was possible using lidar sensors. Equipped with Velodyne's 32-channel lidar, it overcomes the performance limitations of a single sensor by equipping four of the same sensors.

Quanergy Systems has introduced an eight-channel rotational lidar with solid-state scanning. The product is 1/30th the price of Velodyne’s product. It is focusing on mass production of models that secure only a fixed field of view, from motorized rotation to non-rotating lidar products, targeting $250 per unit.

At first, lidar was mounted on the roof of the car due to size issues, but as it became smaller, it is being mounted on the front grill of the car considering design considerations.

ETRI develops independent technology to improve lidar problems

In Korea, the Electronics and Telecommunications Research Institute (KETI) has developed a ‘scanning lidar optical engine platform’ using purely domestic technology, has undergone road driving tests, and is currently developing 360-degree scanning lidar technology. The Electronics and Telecommunications Research Institute (ETRI) has also developed a domestically unique technology called the Static Unitary Detector (STUD) method to complement the inherent problems of existing 3D lidar technology.


This technology is a solution that secures high-resolution 3D image quality at a low cost, and depending on the sensor arrangement structure, it can be used not only for autonomous vehicles but also for drones, surveillance robots, and ultra-small IoT sensors. The developed prototype GEN3 model has a size of 100*100*150mm, which is the same as commercial lidar products. Because it is a stud type, it can be implemented in a wide-angle/360-degree detection form without rotation, the resolution can be freely adjusted even during operation, and it is advantageous for miniaturization.

ETRI said, “LiDAR 3D image sensor technology is changing the fastest and is moving in the direction of requiring low-cost technology,” adding, “This technology is no longer just one of the sensors installed in vehicles, but will play a key role in shaking up the foundation of the future global autonomous vehicle and smart car market.”