Extended Kalman Filter algorithm is used to predict is the position (px, py) and velocity (vx, vy) based on Laser and Radar data provided by car (simulator).

The accuracy of the prediction is calculated using RMSE method.

The steps of the project are as following:

• Read the sensors data and ground truth values from Laser and Radar

• Initialize Extended Kalman Filter

• Predict State Vector (x) and State Covariance (P)

• Update x and P for Laser

• Update x and P for Radar

• Calculate Root Mean Square Error (RMSE) to get the accuracy of estimation against the ground truth values.

The data provided by Laser sensor is in following form:

The data provided by Radar sensor is in following form:

The data provided by server using Extended Kalman Filter is in following form:

The implementation of EKF is divided into 4 files:

main.cpp

fusion_ekf.cpp

kalman_filter.cpp

tools.cpp

Following table summarizes the purpose of each file:

Following points sum up the results and conclusion for EKF:

• EKF is able to handle linear as well as non-linear sensor measurements

• Angle rho normlization is required in Update step in case of Radar measurements

• EKF deals with non-linear measurements by linearizing them using Jacobian Matrix which is not an efficient way of handling non-linearities and losses accuracy in the process

• Unscetned Kalman Filter (EKF) can be used to further reduce RMSE as it handles non-linear measurements using a better mathematical model without needing the linearization step

• Update x and P for Radar

The video (GIF) below shows results of EKF: