Traffic Light and Zebra Crossing Detection Algorithm Based on Bagging Ensemble Learning

The detection of traffic lights and zebra crossings holds significant importance for autonomous driving and ensuring pedestrian safety. Aiming at this issue, a Bagging ensemble learning-based algorithm is proposed in for the detection. The present study establishes a comprehensive dataset comprising traffic lights and zebra crossings, encompassing diverse traffic scenarios such as sunny, cloudy, rainy, and evening. Multiple bootstrap samplings are performed on the training set to obtain multiple sampling sets with different variances and used to train multiple weak learners. The trained weak learners are evaluated on the corresponding out-of-bag datasets, and the detection precisions are taken as their characteristic parameters. The detection results of the weak learners on the test set are integrated, and the reservation or rejection of the detection results is determined according to the respective characteristic parameters and the confidence scores assigned to the boxes. Experimental results show that the ensemble algorithm proposed in this paper can significantly improve the performance of the weak learner and effectively identify traffic lights and zebra crossings.