An improved real-time vehicle detector for autonomous-driving obstacle avoidance, built on the YOLOv8n-seg model with three modifications: the backbone is replaced by FasterNet to cut computational cost and memory while raising detection speed and accuracy; a CBAM convolutional attention module is inserted in the neck to enhance features; and the bounding-box loss is changed from CIoU to WIoU to improve box localization and segmentation accuracy. The result reaches 98.3%, 89.1% and 88.4% accuracy for car, person and motorcycle, outperforming the baseline and YOLOv9 across several metrics.
Computer-vision primary-research paper (Haocheng Guo, Yaqiong Zhang, Lieyang Chen, Arfat Ahmad Khan) on vehicle detection for autonomous-driving obstacle avoidance. Starting from the YOLOv8n-seg mode…