Answers are grounded in this paper and cite their sources.
Different road obstacles may have diverse appearance, size, quantity, and location characteristics, and they are often accompanied by occlusion and overlapping phenomena. These factors result in the current object detection algorithms performing poorly in terms of detection accuracy and robustness when dealing with multiple objects and small objects with indistinct features. To address this issue, a lightweight road obstacle detection algorithm based on the improved YOLOv8n is proposed. Firstly, the MFObstacle road obstacle object detection dataset is created, which expands the types and quantities of obstacles for detection. Secondly, in the Neck part of the network, the GLSA (Global-to-Local Spatial Aggregation) module is added to enhance the multi-scale spatial aggregation mechanism of the network and improve the model's sensitivity and recognition ability for objects of different scales. Then, the BiFPN (Bidirectional Feature Pyramid Network) is integrated. BiFPN can dynamically adjust the weights according to the input data, making the network structure simpler and the computational efficiency higher. Finally, MPDIoU is used. MPDIoU takes into account the overlapping or non-overlapping regions, the distance between the center points, as well as the deviations in width and height, and simplifies the calculation process. Compared with the original YOLOv8n algorithm, the mAP0.5 index of this algorithm is increased by 2.2% to reach 85.7%; the number of network parameters is reduced by 30.4% to 4.6MB; and the computational complexity is reduced by 6% to 7.6GFLOPs. This algorithm improves the detection accuracy while reducing the number of parameters and computational complexity, and can be better applied to the task of multi-object road obstacle detection.
No takes yet. Share an insight, caveat, or question.
Sun et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: