This project is based on Course MIT 6.S094: Deep Learning for Self-Driving Cars and published on this Github
The goal is to predict the steering wheel angel from Tesla dataset based on the video of the forward roadway.
python 3.5 + Keras 2.0.2 + Tensorflow 1.0.1 + Jupyter Notebook + opencv3
These models are trained by P2.xlarge 4vCPUs 61GB on aws.
Databases with real-traffic video data captured and extracted 10 video clips of highway driving from Tesla:
-
The wheel value was extracted from the in-vehicle CAN
-
A window from each video frame is cropped/extracted and provide a CSV linking the window to a wheel value.
A snapshot of video frame:
The CSV data format:
ts_micro | frame_index | wheel |
---|---|---|
1464305394391807 | 0 | -0.5 |
1464305394425141 | 1 | -0.5 |
1464305394458474 | 2 | -0.5 |
in which, ts_micro
is time stamp,frame_index
denotes frame number,wheel
is steering wheel angle(Based on horizontal, + is clockwise, - is anticlockwise)
The generated vedio looks like: