Detecting Potholes Using Simple Image Processing Techniques and Real-world Footage

Potholes are a nuisance, especially in the developing world, and can often result in vehicle damage or physical harm to the vehicle occupants. Drivers can be warned to take evasive action if potholes are detected in real-time. Moreover, their location can be logged and shared to aid other drivers and road maintenance agencies. This paper proposes a vehicle-based computer vision approach to identify potholes using a window-mounted camera. Existing literature on pothole detection uses either theoretically constructed pothole models or footage taken from advantageous vantage points at low speed, rather than footage taken from within a vehicle at speed. A distinguishing feature of the work presented in this paper is that a thorough exercise was performed to create an image library of actual and representative potholes under different conditions, and results are obtained using a part of this library. A model of potholes is constructed using the image library, which is used in an algorithmic approach that combines a road colour model with simple image processing techniques such as a Canny filter and contour detection. Using this approach, it was possible to detect potholes with a precision of 81.8% and recall of 74.4.%.