Residual Mask Based on MobileNet-V2 for Driver's Dangerous Behavior Recognition
Haotian Wang · 2019
We introduce a new residual mask into the inverted residual structure in MobileNet-V2, which significantly improved the performance of the original network, with only a minimal number of parameters added. We train our networks on a larger public dataset to detect distract behaviors of a driver. Experiments show that MobileNet-v2 with the proposed residual mask converges faster and achieves better accuracy than the original network.