Target Detection and Classification in Radar Point Cloud with MathWorks

In autonomous driving functions, an accurate perception of the vehicle environment is a crucial prerequisite. Our goal was to benchmark ADAS sensor performance, particularly RADAR with different radio frequency (RF) and field of view (FoV). Subject to weather conditions, RADAR performance is better than other sensors (camera and ultrasonic) in detecting the target and its attributes (range, velocity, and dimensions). RF signals are transmitted and received by the RADAR mounted on the vehicle frames (front, rear, and side) to enable synchronization and data coherence from all antennas. These signals are modulated in terms of frequency and time domain. The signals are received as point clouds, which are used for target detection and classification. The point clouds are grouped using clustering technique, and attributes of all the points are obtained to develop classification models. We use the Classification Learner app to choose the best classification models to decode ADAS-specific use cases. To have the best confidence factor for the primary target classes such as cars, trucks, motorcycles, bicycles, and pedestrians, we train our classification models (algorithms) based on real-time field data to establish ground truth (GT) information. Our goal is to have an optimized confusion matrix which minimize the false positive rate (FPR) and improve the classification accuracy in different infrastructures, RADAR mounting patterns, and weather conditions. It also enables us to benchmark RADAR performance and determine the ADAS sensor configuration suitable to Renault-Nissan vehicle lines.