Yolov5 Outdoor Dynamic Object Detection Based on Multi-scale Feature Fusion

Outdoor object target detection is a very popular research task. 7 categories of data that fit the outdoor scenario were selected from the VOC2012 dataset as the current dataset for this study. In order to improve the detection accuracy of outdoor objects while keeping the network in good real-time, we propose a new yolov5 outdoor object detection network structure based on multi-scale feature fusion, which fuses shallow and deep features at different scales so that the fused feature layer has rich semantic information while enhancing the positioning information. The MAP of its proposed network on the VOC2012 dataset improved by 1.5% compared to Yolov5, and the FPS reached 61.51, which is better than the original Yolov5 method.