Vehicle Classification and Tracking Using Convolutional Neural Network Based on Darknet Yolo with Coco Dataset

Doğan A., Okatan A., Çetinkaya A.

International Conference on AI and Big Data in Engineering Applications (ICAIBDEA 2021), İstanbul, Turkey, 14 - 15 June 2021, vol.1, pp.179-192

  • Publication Type: Conference Paper / Full Text
  • Volume: 1
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.179-192
  • Istanbul Gelisim University Affiliated: Yes


In this study, vehicle classification based on convolutional neural network has been examined with fully connected layers to apply on highways. With the convolutional neural network, the classification of the vehicles in the image has been made through video images. The accuracy of the classification has been calculated. Convolutional neural networks are multilayer neural networks. Each layer consists of many layers that can be trained. This allows the network to learn better. As the number of layers gets higher the time needed to train the network increases. An open-source Yolo algorithm has been used to provide rapid vehicle detection via video streams. Thanks to this study, the classification of the vehicles present in the inner city and highways is made. Also, the vehicles in the same class are counted. In order to increase the success rate in the experiments carried out, the coco data set is used. In the study, experiments are carried out with video images of 4 vehicle classes taken over 7 different media. 78.75% success rate and 21.25% error rate in the working truck class, 78.08% success rate and 21.92% error rate in the bus class, 80.70% success rate and 19.30% error rate in the motorbike class, car class It has a success rate of 86.21 and an error rate of 13.79%.