Carla_IMU_Classifier

IMU (Inertial Measurement Unit) sensor data reciever & classifier for CARLA SIMULATOR. In both script, Town03 map chosen. Random spawn at start disabled. You can enable with uncomment this line:

spawn_point = random.choice(spawn_points) if spawn_points else carla.Transform()

BE CAREFUL

You have to build Carla simulator before use this repo.

After build, you just have to copy the scripts inside {PATH}\CARLA\PythonAPI\examples folder.

Also, do not forget to execute carla server where {PATH}\CARLA\CARLAUE4.exe with -carla-server argument.

data_reciever.py

You can collect IMU (or other sensors depends on your configuration) sensor data via this script. A new argument "name" created, which is the name of driver.

You can use like that:

python data_reciever.py --filter name --name John29

The script will save a file named "out_John29.csv" which includes 6-axis IMU data with a label column fulled with John29

INSTANCE OUT:

class	accelX	accelY	accelZ	gyroX	gyroY	gyroZ
John29	-0.329013	1.111466	9.943973	0.064446	-0.0759	-0.095295
John29	-0.329013	1.111466	9.943973	0.064446	-0.0759	-0.095295

An example csv file uploaded here ==> examples/out_mehdi_test.csv

classifier.py

You can load & make prediction with trained neural network (tensorflow, .h5 format). The input size of my neural network is (1,20,6)=(Batch, timesteps, features) in my implementation. So, in the line 1074th

data = np.array(data).reshape(1,20,6)

I reshaped data into designed neural network's input size. You can change this line for your configuration.

Requirements

• Python (3.6-3.7)
• tensorflow
• pandas
• pygame
• carla
• numpy

NEWS -> IMU Data reciever for AIRSIM Added!!!

Just copy "car_reciever.py" to the path of path_in_ur_system\AirSim\PythonClient\car and collect IMU DATA!