Earticle

초록

영어

This research explores the utilization of ConvNext Large, a state-of-the-art convolutional neural network architecture, for improving the accuracy and efficiency of car crash detection. By leveraging deep learning techniques, ConvNext Large can automatically extract complex features from raw data, enabling real-time detection of car crashes. Additionally, an active learning framework is proposed to iteratively enhance ConvNext Large's performance by selecting informative unlabeled data points for annotation and inclusion in the training set. The effectiveness of ConvNext Large is compared with other models, including Violent flow, 3D convolution , Updated ResNet\_50, and DenseNet169, highlighting its advantages in handling image-based tasks like car crash detection. Evaluation metrics such as Detection Rate and False Alarm Rate are utilized to assess the accuracy and reliability of car crash detection systems.

목차

Abstract
1. Introduction
2. Related works
3. Methods
3.1. Dataset
3.2. Experiment setup
4. Experiment result
5. Conclusions
Acknowledgement
References

저자

• Ahmed Raza Mohsin [ Dept. of AI Convergence Network Ajou University ]
• Maira Khalid [ Dept. of AI Convergence Network Ajou University ]
• Byeong-hee Roh [ Dept. of AI Convergence Network Ajou University ] Corresponding Author

참고문헌