LIGHTWEIGHT PEDESTRIAN DETECTION ALGORITHM BASED ON ATTENTION MECHANISM

Aimed at the problem of many parameters and low detection accuracy of pedestrian detection model in traffic scene, a lightweight pedestrian detection algorithm integrating attention mechanism is proposed. The model lightweight processing was carried out on YOLOv5 with reference to the Ghost idea. The image hybrid enhancement algorithm was integrated in the data processing part, and coordinate attention (CA) was embedded in the feature extraction network to improve the pedestrian detection accuracy. The regression optimization loss was improved function to improve training speed and inference accuracy. The improved algorithm was tested on the processed Caltech pedestrian dataset. The results show that the average detection accuracy (IOU=0.5) of the improved algorithm is 81.7%, which is 4.4 percentage points higher than that of YOLOv5, and the model parameters are reduced by 45.1%, 3.9×10⁶.