【Industry Trends Forum】Introduction of Elan’s Smart Cockpit & ADAS Solutions

(重點摘要) Edge AI搭配上客製化的訓練用資料集 方能開發出客製化的ADAS與智慧座艙產品
義隆電（ELAN）總經理特助葉宗穎在專題演講中表示，對於ADAS與智慧座艙產品研發，有七個關鍵AI影像技術是要特別注意的，包括：高畫質影像、Edge AI演算法、訓練用資料集、領域知識、AI開發系統、線控技術、實際道路資料。以義隆電的ADAS AI演算法與資料集技術研發來說，2021年義隆電與中央研究院合作，開發出全世界最快且精準度高的物件偵測演算法YOLOv4，之後陸續開發出YOLOR與YOLOv7，並且透過各種實際交通影像測試，不僅能偵測各種不同規格車輛，並且能將辨識結果跟交通號誌控制做結合，自動調整紅綠燈秒數配置，甚至偵測到救護車的時候，交通號誌主動改以讓救護車能快速通過路口需求進行調整。
葉宗穎指出，有了AI演算法之後，廠商在開發上就必須要知道如何挑選正確地資料集進行訓練，甚至透過VR方式產生可訓練用的影像資料集，加速訓練AI辨識模型，並透過VR與AR的方式進行AI訓練，以提升AI模型可用性。由於不同國家在交通環境上有不太一樣的特性，因此義隆電透過機器學習操作（MLOps）方式，依照客戶不同國家交通工具特性需求，進行AI模型客製化訓練。
在智慧座艙方面，葉宗穎認為要針對實際使用情況進行客製化設計，才能符合客戶需求。舉例來說，日本司機習慣戴手套，智慧座艙螢幕觸控就要能夠在使用者戴手套的情況下，能正確感應，在ADAS功能方面，義隆電主要是以Local Edge AI辨識為主，在車用鏡頭解析度720P搭配1 TOPS算力的情況下，可以提供車道偏移警示系統（LDW）、前方碰撞警示系統（FCW）、車距監視與警告系統（HMW）、行人偵測警示系統（PDW）、駕駛狀態監控（DMS）等功能，並且透過支援雷達感測器進行感測器融合（Sensor Fusion）運算，搭配汽車線控系統，可提供L2+等級的ADAS功能。另外，義隆電也針對大巴士使用情境推出對應的L2 ADAS功能，供客戶挑選使用。
------------------------------------------------------------------------
COMPUTEX TAIPEI will return from June 4 to 7, 2024 with the new addition of the Smart Mobility Area. To promote this update, the COMPUTEX Team organized the Auto IC, Driving the Future forum during SEMICON Taiwan on September 7, 2023.
One of the speakers joining the forum is Mr. Joe Tsung-Ying Yeh, Executive Assistant of Elan Microelectronics. Elan Microelectronics is a world leader in human-machine interfaces, notably in Touchscreen Controller with Pen Supports, Touchpad Module, Pointing Stick and Biometrics.
In his speech, Mr. Yeh stated that smart cockpits and ADAS systems will rely on AI image recognition with 7 key technologies including high-quality image, edge AI algorithm, application datasets, specialized & domain knowledge, AI development system, drive-by-wire technology, and practical road experience.
AI models need proper training in terms of quantity, quality, and variety as lack of training will increase risk of collision. Cars equipped with smart cockpits and ADAS systems will need to have their algorithms and datasets continuously updated through OTA updates to make sure the AI knows what to do in various situations. While utilizing live data is generally assumed to be the best way to train AI models, supplementing the datasets with simulated data and training through AR/VR or even generative AI can further help improve the model.
COMPUTEX will return from June 4 to 7 2024 in Taipei Nangang Exhibition Center (TaiNEX) Halls 1 and 2. Join us onsite and find the latest ICT products in Asia’s leading B2B ICT Trade Show!
For more details, please visit
Official Website : www.computex.biz/
Online Exhibition Channel : show.computex.biz/online.aspx ( COMPUTEX CYBERWORLD )
Follow us on social media for more news and updates:
Facebook: / computextaipei
LinkedIn: / computex-taipei
#computex #COMPUTEX2024 #computexcyberworld