Early intent prediction of vulnerable road users from visual attributes using multi-task learning network | IEEE Conference Publication | IEEE Xplore