TRAFFIC CONGESTION DETECTION THROUGH OBJECT DETECTION APROACH USING FASTER R-CNN AND SINGLE SHOT DETECTOR FRAMEWORK

• Main Author: Arjuna Purba, Regi
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50196
• Institution: Institut Teknologi Bandung

Traffic congestion is a main problem that lead to serious impact such as material loss, psychological illness, and loss of time. Object detection is a problem to localize and classify objects in picture or video. Object detection can be use as an approach to solve traffic congestion. This can be done using object detection to get information about traffic density in traffic road. Object detection model can be used to detect and classify objects that causes congestion. The problem is how the model could detect objects and classify objects on input data and give output about traffic condition as CONGESTED or NORMAL. Faster R-CNN and Single Shot Detector Framework is framework to do object detection. This framework is based on Convolutional Neural Network architecture. In this research use Inception and Residual Network architecture. Through the design and implementation the result show that Faster R-CNN with the Inception architecture is the best model to detect object. This result is supported by the fact that this model have the highest accuracy among the other chosen model. This model get 0.88 of accuracy when the traffic is congested and 0.82 of accuracy when the traffic is normal. But, Single Shot Detector Framework get the fastest inference time. Single Shot Detector get 0.97 second of inference time when the road is congested and 0.87 second of inference time when the road is normal. Faster R-CNN framework with the Inception architecture is used as a model to detect traffic congestion. The strategy to classify traffic condition use number of object detected threshold and structural similarity threshold. From the implementation of the designed strategy show that the program is successful to classify traffic condition as CONGESTED and Normal with 0.75 of accuracy