A lightweight object detection algorithm for traffic scenes based on an improved YOLOv7-tiny

In response to the challenges that traditional image processing algorithms fail to detect traffic-scene objects accurately and efficiently, and that existing deep learning–based detection models often suffer from high computational complexity and low inference speed, this paper proposes a lightweight object detection algorithm based on an improved YOLOv7-tiny architecture. The proposed method integrates the Spatial and Channel Reconstruction Convolution (SCConv) module to effectively suppress feature redundancy, thereby significantly reducing model parameters and computational overhead while enhancing the feature representation capability of the network. Furthermore, a Convolutional Block Attention Module (CBAM) is embedded after the ELAN block in the backbone to strengthen the extraction of key feature information. In the neck network, the original PANet is replaced with a Bidirectional Feature Pyramid Network (BiFPN) to improve multi-scale feature fusion and spatial information representation. Experimental results on a subset of the BDD100K autonomous driving dataset demonstrate that the proposed method achieves approximately a 50% reduction in model parameters and computation while improving the mean Average Precision (mAP) by 0.02. The improved network achieves a better trade-off between accuracy and efficiency, making it more suitable for real-time object detection in complex traffic scenarios.