Research on Complex Street View Detection Algorithm Based on YOLOv8

The current street scene is becoming more and more complex, with high pedestrian and vehicular traffic and a large number of targets. Aiming at the problems that the accuracy of the current complex street scene-oriented detection algorithms is not satisfactory and the leakage rate of small targets is unstable, the intelligent detection algorithm YOLOv8s- for complex street scenes is proposed, aiming at the intelligent detection of multi-targets in daily street scenes. This study is based on the target detection model of YOLOv8s, proposing a C2f-DCNv2 module that can sample features flexibly; adding a small target detection layer in the head; and using the Shape IoU loss function. The improved algorithm improves the precision, recall, and average accuracy by 2.2%, 3.6%, and 5.0%, respectively, compared with the initial algorithm, and these experimental data validate that this study can have a good performance when targeting the detection task of complex street scenes.