Skip to main content
:::
  • HOME
  • News
  • College of Electrical Engineering and Computer Science, NCHU Wins Future Technology Award: LiDAR Technology Integration & AOI Intelligent Optical Inspection Image Component Detection

College of Electrical Engineering and Computer Science, NCHU Wins Future Technology Award: LiDAR Technology Integration & AOI Intelligent Optical Inspection Image Component Detection

2022-01-18 09:00:08
興新聞張貼者
Unit秘書室
308
The key to the popularization of autonomous vehicles lies in "technological breakthroughs" and "cost reduction". Among them, one of the most critical environmental sensing technologies is "LiDAR", which must be able to detect "objects at medium and long distances" and "identify the contour and speed of objects" at the same time. However, the current mechanical LiDAR, which is placed on the roof of the car and rotating, is not only inconsistent with the aesthetics of the car's craftsmanship, but also has the problems that dustproofing, waterproofing and electrical system installation are not easy. In response to this problem, the team from National Chung Hsing University College of Electrical Engineering and Computer Science has developed "Embedded LiDAR and AI Recognition of Smart Headlight for Autonomous Driving", using solid-state embedded 1550nm wavelength LiDAR, and combined with AI recognition of smart headlight. Autonomous vehicles based on point cloud and image fusion data provide "environment recognition", "distance and speed" information and quickly give feedback to the smart headlight control system, which can switch the most suitable headlight type under different road conditions to meet the needs of safe driving. The biggest breakthrough of this technology is that the current technology is relatively mature, and the LiDAR system’s light source with 905-nm wavelength may be harmful to the human eye. However, the 1550-nm long wavelength light source has high power, reduces background noise, and increases the sensing distance and accuracy rate.

In autonomous vehicles, LiDAR sensing and headlight control are the key. The team from National Chung Hsing University College of Electrical Engineering and Computer Science uses solid-state LiDAR embedded in the headlights, and meanwhile, with the "adaptive driving beam system" (ADB), the team’s unique core single crystal phosphor and blue laser are used to provide high-brightness light source. With digital micromirror device, this smart headlight can quickly change the "autonomous vehicle lamp lighting type" (high beam and low beam switching, avoiding high reflective areas, strengthening weak light areas, lighting warnings, forecasting driving directions and two-way communication with the ground projection and other functions), adapt to different driving conditions (including overtaking in curves) and two-way communication between pedestrians and vehicles on the ground, while complying with the lamp safety regulations and standards (ECER112 and SAE J3069), and achieving the requirements of autonomous and safe driving. This technology has been tested in cooperation with major domestic headlight manufacturers. Foreign smart headlight competitors have not yet presented related applications and technologies. Enhancing international competitiveness, Taiwan’s headlight industry will establish a firm position as a Tier 1 supplier. It will become even more powerful.

As the Principal Investigator, Professor Mu-Hai Cheng, pointed out, the traditional car’s mechanical LiDAR is large and costly as high as 70,000 US dollars. In the future, lightness, thinness, shortness and smallness will be the trend, and reducing the cost to 1,000 US dollars is the goal they are striving for. The technology of the “Embedded LiDAR of 1550 nm Wavelength and AI Recognition System” developed by the team from National Chung Hsing University College of Electrical Engineering and Computer Science is far ahead of China’s LiDAR technology in ranging accuracy and time synchronization, but it is equally competitive in terms of cost of manufacture. With the estimated output value of ADB smart headlights of NT$50 billion, the prospects are promising. The technological breakthrough also won the honor of 2021 Future Technology Award from the Ministry of Science and Technology. The team also hopes that through innovative and breakthrough technologies, it will assist Taiwan’s traditional headlight industry and local companies in formulating smart headlight industry strategies, patent layouts and related supporting measures, vertically integrating the upstream and downstream headlight supply chains, and promoting them to terminal automotive manufacturer customers. Then we can promote industrial upgrading and enhance Taiwan's ability to participate and influence in the autonomous vehicle market.

The technology that can achieve the “Embedded LiDAR of 1550 nm Wavelength and AI Recognition System” is based on “1550-nm solid-state LiDAR chip” developed by YuShan Scholar and Professor Charles W. Tu from the Department of Electrical Engineering, NCHU, and Professor Tien-Tsung Shih and Chun-Nien Liu, members of the project team. The key is that the team integrates "OPA (optical phased array)" technology, "1550-nm VCSEL (vertical-cavity surface-emitting laser)" to form "micro solid-state LiDAR" that has a detection distance of up to 100 meters. It is also the technology with the highest scanning rate among the current LiDAR systems (30° x 2.5°, scanning speed 1 kHz), and it is applied to the fourth generation of LiDAR. The team not only acquired Value Creation Program of Ministry of Science and Technology, but also won the 2021 Future Technology Award of the Ministry. As Professor Charles W. Tu further explained, this technology can not only be used in the integration of autonomous vehicle LiDAR systems, but also can fully be employed in the "augmented reality" function of mobile phones. In the future, it can even be combined with smart glasses to make human life more convenient.

In addition, "From the First Small Beginnings One Can See How Things Will Develop: The Detection of Artificial Intelligence Optical Detection Image Component Based on Few Shot Learning" by Professor Chun-Jung Huang from National Chung Hsing University College of Electrical Engineering and Computer Science also won the honor of the "2021 Future Technology Award" of Ministry of Science and Technology. It is expected that this technology solves the problem of "auto-optical inspection" in the process of "deep learning" that requires a lot of data and manual calibration, which does not meet cost considerations and speed.
The deep learning based artificial intelligence (AI) models currently used in the industry require a lot of training data to achieve high accuracy. On the grounds that the deep learning based AI models require long training time and have high computational complexity, they need to rely on expensive graphics processing unit (GPU) equipment to support the huge amount of computation and accelerate model training. Ordinary small and medium-sized enterprises simply cannot afford the cost. The team has developed a few shot self-supervised learning method to solve the problem of "accurately locating target components based on only a few labelled images". Professor Chun-Rong Huang explained that it also solves the problem of consumption of 80% of data processing time in early and later stages, and training data labeling in the "deep learning" process of AI project development in the manufacturing process of the technology industry which conventionally requires a lot of manpower costs.

Professor Chun-Rong Huang and his team have broken away from the traditional deep learning detection models. With a brand-new artificial intelligence algorithm, on the basis of images, and through the "few shot self-supervised learning method", they can perform component detection for different components effectively. The user only needs to drag the cursor to select the position of the target component to complete the labeling, and their method can even work on special polygonal components. Compared with the traditional deep learning methods that require nearly a thousand image samples, the new technology only needs fewer than ten samples to achieve the same as or even higher performance. More importantly, the novel method can be operated on ordinary computers with only CPU equipment. The novel method greatly boosts industrial upgrading, improves the detection efficiency and accuracy of AOI, and increases the learning efficiency of AI models.

This technology has now completed the product inspection and verification of the production lines in major domestic technology companies, which can help increase the yield and the accuracy of product quality control. The team also expects that it can be gradually extended to domestic companies, traditional industries and other production lines that have AOI requirements to help the Taiwan’s industry respond to the rapidly changing competitive market.

Back