A novel real-time object detection method for complex road scenes based on YOLOv7-tiny

Road object detection is a key technology in intelligent transportation systems, playing a crucial role in ensuring driving safety and enhancing driving experience. However, due to factors such as weather and visual occlusions, particularly in complex traffic scenes, the recognition rate and accuracy of object detection are often less than satisfactory, far from meeting the application demands of intelligent driving. In order to address the issues of weak generalization and low regression accuracy of image similarity evaluation metrics, we propose a new anchor box calculation algorithm. Building upon this, to tackle the problem of weak graphic attention and feature capture capabilities in the backbone network,We propose an improved CA attention mechanism. In addition, to address the issues of low detection accuracy and imprecise positioning of the model in complex traffic scenarios, we propose a new image enhancement module. we select the road traffic dataset BDD(Berkeley Deep Drive)100K as the benchmark evaluation dataset and divide the training and validation sets into six new categories. Through this series of strategies, a new real-time road object detection method suitable for complex traffic scenes is formed. To validate the effectiveness of this method, we conducted a series of experiments. The experimental results demonstrate that our proposed method achieves a mean average precision improvement of 3.61% compared to the YOLOv7-tiny method.