Deep Learnt Anti Collision System for Highway Safety

Many accidents occur at highway roads due to the driver's unawareness, about the status of the upright vehicles before them. The lethargic behavior of the driver may lead to improper parking of vehicles on the road (i.e. on road/ Off Road) or the breakdown of vehicle may cause the vehicle to stop on the road which creates risk to the fore coming vehicles. Hence, an automated system is required to alert the vehicles coming close to the static vehicle based on their parking positions on the road. This paper has proposed an image processing algorithm using convolutional neural network to identify the position of standing vehicle from the image captured by its rear end camera. A convolutional neural network is trained with the set of images acquired at "on road parking", "off road parking" and "partially on road parking". The features are learnt by the neural network model and the status of the query image is identified. In this approach, the status about the mobility of vehicle is tracked with the help of tachometer. Based on this, a state model is coined to initiate an RF signal from a standing vehicle to near coming vehicle. The accuracy of the proposed approach is validated with the standard performance measures like precision and recall.