Research Article
Improving Vehicle Speed Detection with Ensemble Learning: A Comparison of CNN Architectures
@INPROCEEDINGS{10.4108/eai.28-4-2025.2358005, author={Garikina Praveen and Ankam Purna Naga Sri and Sidda Yamuna and Mohammed Saniya Fathima Khannam and Mulukuri Siva Sai Sandeep and Tharun Vudugala}, title={Improving Vehicle Speed Detection with Ensemble Learning: A Comparison of CNN Architectures}, proceedings={Proceedings of the 4th International Conference on Information Technology, Civil Innovation, Science, and Management, ICITSM 2025, 28-29 April 2025, Tiruchengode, Tamil Nadu, India, Part II}, publisher={EAI}, proceedings_a={ICITSM PART II}, year={2025}, month={10}, keywords={deep learning ensemble vehicle classification transfer learning computer vision transportation systems real-time detection}, doi={10.4108/eai.28-4-2025.2358005} }- Garikina Praveen
Ankam Purna Naga Sri
Sidda Yamuna
Mohammed Saniya Fathima Khannam
Mulukuri Siva Sai Sandeep
Tharun Vudugala
Year: 2025
Improving Vehicle Speed Detection with Ensemble Learning: A Comparison of CNN Architectures
ICITSM PART II
EAI
DOI: 10.4108/eai.28-4-2025.2358005
Abstract
VSD is one of the most serious issue in ITS since it is indispensable for almost all traffic control and safety systems. An enhanced architecture of VSD was proposed in the current work by integrating deep learning and optical flow techniques. The proposed approach involved pre-processing such as feature extraction that was carried out to alter the model by combining pre-processing layers and scaling the data of the model. Moreover, a strong ensemble model based on InceptionV3, ResNet152V2 and EfficientNetB7 was employed to classify the tested vehicles and to predict their speed. Each model’s output was a prediction of the speed, the predictions were debiased, and then all these speeds were ensembled to enhance the accuracy and stability of the predicted speeds. Lastly, an extensive evaluation process with accuracy, precision, recall, and F1-Score metrics proved to have better prediction time and lower errors. It renders the proposed ensemble-based technique efficient and practical for practical traffic monitoring systems.
