Deep detection network for real-life traffic sign in vehicular networks.

•A new framework for real-life traffic sign detection is proposed which extracts region proposals in a coarse-to-fine manner.•The novel module called attention network is introduced to locate the RoIs coarsely and classify the targets roughly.•A deconvolutional struction is added to convolutional layers to fit the small size of targets, and the final classifier is replaced by three softmax classifiers.