Violation Detection on the Roadside of the Toll Roads using Cascade R-CNN Algorithm

The shoulder of the toll road is the left part of the toll road that is useful for vehicles experiencing emergencies. However, there are still many vehicles that use the shoulder unwisely, such as using it to pass other vehicles or stopping or resting for a long time. This can cause accidents and commotion. Based on the problem, the Cascade R-CNN algorithm is applied which can be used as an algorithm to detect violations on the toll road shoulder. Cascade R-CNN has multistage detection which can reduce overfitting due to lack of dataset. Cascade R-CNN has multistage detection consisting of three stages. The detection train results in the first stage will be trained again in the second stage, and so on until the third stage results. Cascade R-CNN is a well-regarded technique for object detection because of its multistage detection process. Three hyperparameters—epoch, batch size, and learning rate —are used to assess the performance of the Cascade R-CNN algorithm. The goal of hyperparameter testing is to find the most accurate prediction model. The optimal model was created using hyperparameters at epoch 12, batch size 16, learning rate 0.02, Average Precision=97.1%, Average Recall=79.1%, mAP@.5=97.1%, and mAP@.5:.95=74.8%.