Application of Machine Learning Algorithms for Detection and Tracking of Multiple Targets Based on Experimental Data Acquired Using Radar Sensor Operating in the 77 GHz Band

Abstract

Small form factor radar sensors at millimeter wavelengths nd numerous applications in the industrial and automotive sectors. These radar sensors provide improved range resolution, good angular resolution, and enhanced Doppler resolution for short-range and ultra short-ranges. The primary objective of this thesis is to detect and track the targets accurately, when a radar is interfered by another. This research proposes an experimental evaluation of a 77 GHz IWR1642 radar sensor in the presence of a second 77 GHz AWR1642 radar sensor acting as a spot jammer. A real-time experiment is carried out by taking into account ve di erent targets of various radar cross sections, such as a car, a larger size motorcycle, a smaller size motorcycle, a cyclist, and a pedestrian. The collected real-time data is processed using four di erent constant false alarm rate (CFAR) detectors, CA-CFAR (cell averaging-CFAR), OS-CFAR (order statistics- CFAR), GOCA-CFAR (greatest of cell averaging-CFAR), and SOCA-CFAR (smallest of cell averaging-CFAR). Furthermore, the data from the above detectors is fed into two di erent clustering algorithms (DBSCAN (density based spatial clustering of applications with noise) and K-means), followed by the extended Kalman lter (EKF) based tracker with global nearest neighbor data (GNN) association, which provide tracks of various targets with and without the presence of a jammer. Furthermore, four di erent metrics (1.Tracks reported (TR), 2.Track segments (TS), 3.False tracks (FT), and 4.Track loss (TL)) are used to evaluate the performance of various tracks generated for two clustering algorithms with four detection schemes. The experimental results show that the DBSCAN clustering algorithm outperforms the K-means clustering algorithm for many cases. In addition, deep learning (DL) based technique is applied for the range Doppler (RD) maps obtained from 77 GHz mmWave sensor. CA-CFAR and OS-CFAR detections are labeled by LABELIMG software, and then these images are processed by you only look once (YOLO) V5 method. The results obtained have been compared with ground truth information for various cases. The results obtained reveals that, the DL based approach provides good performance in terms of mean absolute average error. Due to the increased on-road density of mmWave radars, the primary radar mounted on the ego vehicle faces mutual interference. As a result of this, another contribution is made in the thesis to mitigate the mutual interference. To address this interference problem, two novel algorithms are proposed. One of the method is weighted beat signal normalization and the other is clipping followed by a hampel ltering algorithm. Both these approaches are applied on IF (intermediate frequency) signal followed by a traditional detection scheme as a mutual interference mitigation mechanism. Finally, simulated clutter is generated to see the target detection using a novel sigma delta space time adaptive processing ( STAP) technique to detect multiple targets in the presence of clutter.