Speed Breaker Identification Using Deep Learning Convolutional Neural Network

Abstract

AbstractSpeed breakers are one of the major reasons for road accidents in recent years. Speed breakers are constructed for human safety nearby schools and hospitals. Improper dimensions, absence of signboards, and unmarked speed breakers are the major threat to accidents. Real-time speed breaker detection is important to avoid road accidents, and all autonomous vehicles need to have this facility. Real-time identification of speed breakers is important and also difficult due to their different dimension and colour. The existing speed breaker detection method uses the accelerometer, GPS geotagged smartphone and image processing technique. Sensor data vibration pattern, GPS error, network overload and battery depletion are some of the shortcomings of these technologies. Due to these drawbacks, these approaches cannot be used to identify road conditions in real-time. In this work, Deep learning neural network is used for the detection of speed breakers. For a range of computer vision and medical image processing tasks, deep learning neural networks have proven to be effective. This research provides a new architecture for a convolutional neural network (CNN) based speed breaker detecting system. In CNN, the detection accuracy is increased by creating a bounding box with a localization approach. The CNN is trained on a library of speed breaker photos that have been gathered over time. The trained CNN is utilised to identify speed bumps in real-time. Prediction accuracy of 55% is obtained for real-time data by the trained network.KeywordsAccidentSpeed breakerConvolutional neural networksDeep learning