Pothole Detection System (PoDS) with Computer Vision and Internet of Things (IoT)

Pothole is one of the key factors that lead to road accidents in Malaysia each year. In this paper, a Pothole Detectio n System (PoDS) is developed where machine learning algorithm, computer vision and Internet of Things (IoT) play an important role in detecting potholes. The project is divided into two parts, the first part is focused on object detection with the application of YOLO (You Only Look Once) algorithm which involves labelling of pothole dataset, training the pothole detection AI model through Google Colab (Google Colaboratory) and testing the pothole detection AI model. For the computer vision to work in this system, NVIDIA Jetson Nano Developer Kit, a single board computer, is connected with a web camera to detect the potholes. Another part of this project involves the IoT implementation where a GPS module is used to identify the pothole locations in real time whereas NodeMCU functioned as the microcontroller along with a Wi-Fi module. These two components are integrated in order to obtain the location data of potholes and which is then saved into the ThingSpeak cloud database. Field tests are carried out to test the general performance of the PoDS system. The results obtained are very positive as it detected most of the observable potholes. The success rate under different situations was studied and it was found that the detection success rate during daytime was 94.74%. Also, the detection of rainwater-filled potholes had a success rate of 71.43% while detection during night time (illuminated by headlamp and streetlight) had a success rate of 43.75%. With the ability to detect potholes, it may help the road users especially motorcyclists to avoid bumping into potholes and prevent life-threatening road accidents. Furthermore, the data for pothole locations can be used for further developments in different industries such as the automotive industry.

Pothole is one of the key factors that lead to road accidents in Malaysia each year. In this paper, a Pothole Detectio n System (PoDS) is developed where machine learning algorithm, computer vision and Internet of Things (IoT) play an important role in detecting potholes. The project is divided into two parts, the first part is focused on object detection with the application of YOLO (You Only Look Once) algorithm which involves labelling of pothole dataset, training the pothole detection AI model through Google Colab (Google Colaboratory) and testing the pothole detection AI model. For the computer vision to work in this system, NVIDIA Jetson Nano Developer Kit, a single board computer, is connected with a web camera to detect the potholes. Another part of this project involves the IoT implementation where a GPS module is used to identify the pothole locations in real time whereas NodeMCU functioned as the microcontroller along with a Wi-Fi module. These two components are integrated in order to obtain the location data of potholes and which is then saved into the ThingSpeak cloud database. Field tests are carried out to test the general performance of the PoDS system. The results obtained are very positive as it detected most of the observable potholes. The success rate under different situations was studied and it was found that the detection success rate during daytime was 94.74%. Also, the detection of rainwater-filled potholes had a success rate of 71.43% while detection during night time (illuminated by headlamp and streetlight) had a success rate of 43.75%. With the ability to detect potholes, it may help the road users especially motorcyclists to avoid bumping into potholes and prevent life-threatening road accidents. Furthermore, the data for pothole locations can be used for further developments in different industries such as the automotive industry.