Research on the object detection algorithm with modified YOLOv7 for autonomous driving applications

Abstract

Aiming at the problems of low recognition accuracy, missed detection, and false detection of targets in autonomous driving scenarios of the original YOLOv7, an improved YOLOv7 object detection algorithm is proposed. In the improved algorithm, the SimAM attention mechanism is first introduced at the neck of the network, which can evaluate feature weights in the 3D dimension without adding additional parameters to enhance important features, suppress invalid features, and achieve high detection speed while meeting accuracy requirements. Secondly, choosing EIoU Loss helps the model better learn multi label classification tasks, improving the performance and generalization ability of the model in cases of imbalanced and overlapping labels. The experimental results show that the Precision (P) of the improved algorithm is 75.3%, and the Recall rate (R) is 56.6%. The mean Average Precision (mAP) reached 54.2%. The improved algorithm demonstrates good performance in real-scene detection tasks, effectively reducing the missed detection rate and false detection rate, while significantly improving the detection ability and accuracy of the model.