YOLOv7-Based Autonomous Driving Object Detection Algorithm

This article introduces an enhanced YOLOv7 algorithm for automatic driving object detection, aiming to address the diverse road scenarios, missed detections, and false alarms encountered in automatic driving object detection. Firstly, the article adopts the WIoU loss function to prioritize the average quality of anchor boxes. Secondly, the article utilizes the CARAFE upsampling operator to enhance the model's ability to accurately reassemble feature maps. Lastly, through the introduction of the CBAM fusion attention mechanism to optimize the backbone network ELAN module, the algorithm effectively mitigates background interference and improves detection accuracy by appropriately allocating network weights in spatial and channel aspects. Experimental results on the KITTI dataset demonstrate that the improved YOLOv7 algorithm performs exceptionally well in object detection, achieving an accuracy of 94%, a recall rate of 89.8%, mAP@0.5 of 94.8%, and mAP@.5:.95 of 71.2%. Compared to the original YOLOv7 algorithm, these improvements represent an increase of 2.1%, 2.5%, 2.7%, and 3.8% respectively, clearly demonstrating the effectiveness of the improved algorithm.