With the continuous development of artificial intelligence and computer vision technology, autonomous vehicles have developed rapidly. Although self-driving vehicles have achieved good results in normal environments, driving in adverse weather can still pose a challenge to driving safety. To improve the detection ability of self-driving vehicles in harsh environments, we first construct a new color levels offset compensation model to perform adaptive color levels correction on images, which can effectively improve the clarity of targets in adverse weather and facilitate the detection and recognition of targets. Then, we compare several common one-stage target detection algorithms and improve on the best-performing YOLOv5 algorithm. We optimize the parameters of the Backbone of the YOLOv5 algorithm by increasing the number of model parameters and incorporating the Transformer and CBAM into the YOLOv5 algorithm. At the same time, we use the loss function of EIOU to replace the loss function of the original CIOU. Finally, through the ablation experiment comparison, the improved algorithm improves the detection rate of the targets, with the mAP reaching 94.7% and the FPS being 199.86.

Convolutional neural networks have achieved good results in target detection in many application scenarios, but convolutional neural networks still face great challenges when facing scenarios with small target sizes and complex background environments. To solve the problem of low accuracy of infrared weak target detection in complex scenes, and considering the real-time requirements of the detection task, we choose the YOLOv5s target detection algorithm for improvement. We add the Bottleneck Transformer structure and CoordConv to the network to optimize the model parameters and improve the performance of the detection network. Meanwhile, a two-dimensional Gaussian distribution is used to describe the importance of pixel points in the target frame, and the normalized Guassian Wasserstein distance (NWD) is used to measure the similarity between the prediction frame and the true frame to characterize the loss function of weak targets, which will help highlight the targets with flat positional deviation transformation and improve the detection accuracy. Finally, through experimental verification, compared with other mainstream detection algorithms, the improved algorithm in this paper significantly improves the target detection accuracy, with the mAP reaching 96.7 percent, which is 2.2 percentage points higher compared with Yolov5s.

No abstract