夜間長距離AI辨識技術

應用毫/微米級LED之AI智慧頭燈開發計畫

運用AI夜間影像辨識技術，透過夜間影像資料的蒐集、標註與模型訓練，使模型具備有效的特徵提取能力，並以夜間及遠距離車輛影像資料集為基礎，進行模型架構調整與參數優化，透過模型特徵結構的調整，改善影像特徵在各層之間的傳遞與整合，使模型能更準確地提取夜間車輛輪廓與光源特徵，另外，針對遠距離、光暈干擾、低亮度場景、及小物件進行微調訓練，以提升模型在低亮度與小目標物件下的辨識表現，加強模型對真實夜間情境中的辨識穩定性與準確度。

This AI-based night-time image recognition technology optimizes model architecture and parameters using annotated night-time and long-distance datasets. By refining cross-layer feature integration, the model accurately captures vehicle contours and light sources. Furthermore, targeted fine-tuning addresses challenges like long-distance detection, glare, low light, and small objects, significantly improving recognition stability and accuracy in real-world nighttime environments.