This technology uses C3D-based deep learning network, entering rear camera image to detect rear overtaking behavior.  C3D deep learning architecture is different to 2D one by entering not only 1 image but continuous 16 images. Therefore, using this network can directly learn target object behavior,and know what  overtaking is. It can be applied in E-mirror products to ensure safer driving.

This technology is modified from C3D network. With the limitation of low resolution image, not only classifying the behavior, but showing the heat-map represented the overtaking behavior by adding few convolution computation are included. It achieve more then 95 precision rate at daynight. Compared to 2D CNN, it is first system that can detect the objectits behavior in the world.

This system can be applied into ADAS, autonomous car, e.g., overtaking, pedestrian passing, forward car suddenly braking.