OHDL: Radar target detection using optimized hybrid deep learning for automotive FMCW

The risk of collision increases, as the number of cars on the road increases. Automotive radar is an important way to improve road traffic safety and provide driver assistance. The FMCW radar principle is extensively utilized in vehicle radar systems. FMCW-Radars operate by generating a transmit signal that consists of a linear frequency ramp. The information obtained from a single frequency ramp is insufficient to accurately determine both range and velocity simultaneously due to its inherent uncertainty. In order to eliminate the uncertainty between the frequency component generated by range and the Doppler frequency, it is necessary to generate multiple consecutive ramps. In this paper, we proposed a novel optimized hybrid deep learning model (OHDL). The radar echo data cube is used to extract the deep features and performed the range-Doppler (RD) map prediction and range and velocity detection. In this for hyper parameter optimization in deep learning model, we used new enhanced particle swarm optimization (PSO) algorithm. The performance evaluation is implemented using different scenarios of simulation and compared with earlier works of radar target detection (RTD). Root mean square error (RMSE) is nearly 50% reduced than the existing methods in our OHDL.