Road Object Detection in Bangladesh Using Faster R-CNN

Abstract

The importance of object detection in our lives is increasing day by day. The role of object detection is very important in autonomous cars, intelligent driving assistance, and advanced traffic analysis. In the case of traffic analysis and intelligent driving assistance in Bangladesh, it is very important to properly identify all the objects from real-time video. Because in both cases the main responsibility of the system is to give the driver or authority a clear idea about the road or the environment around the vehicle. And for this, we need to use modern algorithms and architecture based neural network models with much better object detection accuracy such as Faster R-CNN. There are currently a couple of algorithms that work faster than Faster R-CNN but cannot detect objects accurately, as is the case with small-to-medium objects. We used Faster R-CNN on our data to analyze the environment around the road and the environment around the car. We trained the network for 19 object classes and tested its ability to detect objects with real-time video analysis with an accuracy of 86.42%. Moreover, FPR(false positive rate) and FNR(false negative rate) is calculated to evaluate the proposed model from confusion matrices. In this study, the FPR of the Faster R-CNN model is 15.97% and the FNR of the Faster R-CNN model is 12.2%