SLDNet: A Branched, Spatio-Temporal Convolution Neural Network for Detecting Solid Line Driving Violation in Intelligent Transportation Systems

摘要

Solid line driving is known as one of the major driving violations in China. In this paper, we propose a branched, spatio-temporal convolution neural network, named SLDNet, to recognize these violation acts from photographs captured by surveillance cameras and train it on Pingxiang solid-line-driving dataset. SLDNet can achieve 0.91 in accuracy and 0.92 in recall, out-perform the current human review (80% in accuracy). Our method will be implemented in intelligent transportation systems in Pingxiang city, Jiangxi Province in near future.