Near-Field SAR Imaging of Moving Targets on Roads
Raz Rajwan, Amir Boag, Fellow, IEEE · 2025
Read Near-Field SAR Imaging of Moving Targets on Roads by Raz Rajwan, Amir Boag, Fellow, IEEE (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-18579.txt.
Introduction
Raz Rajwan and Amir Boag image moving vehicles on roads with a single radar channel flying a circular path, in this 2025 paper. Near-field radar in spotlight mode can watch traffic, but movers smear under standard processing. Their algorithm is built for exactly that case.
A fast back-projection scheme stitches together coarse images from slices of the data, using layered interpolation to stay quick. Both parked and moving vehicles come out sharp. The design assumes the radar circles the scene, a pattern drones and towers can fly. The authors present it as a step toward cheap, high-fidelity traffic radar.
Catalog record
| CREATOR | Raz Rajwan, Amir Boag, Fellow, IEEE |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-18579
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Near-Field SAR Imaging of Moving Targets on Roads Raz Rajwan, Amir Boag, Fellow, IEEE Abstract This paper introduces a single-channel SAR algorithm designed to detect and produce high-fidelity images of moving targets in spotlight mode. The proposed fast backprojection algorithm utilizes multi-level interpolations and aggregation of coarse images produced from partial datasets. Specifically designed for near-field scenarios and assuming a circular radar trajectory, the algorithm demonstrates enhanced efficiency in detecting both moving and stationary vehicles on roads. 0 0 footnotetext: This work was supported in part by the BSF under Grant 2022346 and in part by the NATO SPS under Grant G6118. The authors are with the School of Electrical and Computer Engineering, Tel Aviv University, Tel Aviv, 69978, Israel (e-mail: [email protected]; [email protected]). I Introduction Synthetic Aperture Radar (SAR) has emerged as a powerful tool for remote sensing applications, offering high-resolution imaging capabilities regardless of weather conditions or time of day. SAR imaging is based on coherent processing of scattered signals over a range of frequencies and angles. The reflected signal can be modeled as a convolution between the pulse waveform and the ground reflectivity function [ 1 ] . A typical approach to SAR imaging involves the numerical evaluation of an integral operator [ 2 ] . The integral is often formulated as a multidimensional Fourier transform allowing for efficient evaluation using Fast Fourier Transform (FFT) based algorithms [ 3 ] and [ 4 ] . Utilizing the FFT lowers the computational cost of computing a two-dimensional N\times N image from O(N^{4}) to O(N^{2}\log N) , where N denotes the number of samples in both the range and azimuth dimensions. However, the reduction of the imaging operator to FFT form is frequently achieved by far-reaching approximations, which cause degradation of the image quality. The common methods for SAR image formation include the Polar Format Algorithm (PFA) [ 5 ] , which operates under the assumption of the far-field approximation. Additionally, the Range Migration Algorithm (RMA) [ 6 ] presumes uniform sampling in both azimuth and range dimensions. The Chirp Scaling Algorithm (CSA) [ 7 ] excludes interpolation but does not provide complete compensation for the spherical nature of the waves. A comprehensive comparison of these algorithms is given in [ 5 ] and [ 8 ] . Several fast backprojection (BP) algorithms have been proposed in [ 9 , 10 , 11 , 12 , 13 , 14 ] . These algorithms demonstrate greater imaging quality and flexibility when compared to the FFT-based ones, while retaining the same CC. Here, we propose an improved algorithm for a single-channel radar by extending the divide-and-conquer approach proposed in [ 9 ] and [ 15 ] , with our focus on the near-field, spotlight mode scenario. In these papers, a recursive multi-level algorithm is introduced to efficiently compute the BP operator over the 4-D space. The adoption of the multi-level BP image formation is preferred due to its high accuracy, robustness, and flexibility, particularly in compensating for variations in antenna patterns across different angles and frequencies. Moreover, the BP algorithm has the capability to overcome near-field limitations that are often encountered in other algorithms. In this study, we propose several modifications to improve the CC of the multi-level algorithm. Firstly, we identify roads within the static image. Subsequently, we demonstrate the feasibility of implementing the multi-level algorithm solely on a 2-D space comprising the detected roads, resulting in a significant reduction in complexity. Additionally, we adapt the multi-level algorithm to the near-field scenario and examine the limitations of this method. The article is organized as follows: The BP operator for moving targets in the near field scenario is formulated in Sec. II . The 4-D multi-level algorithm and the adaptive algorithm for detecting and imaging moving targets are introduced in Sec. III . An improved algorithm for detecting and imaging targets on roads is presented in Sec. IV . The algorithm’s performance is demonstrated and analyzed in Sec. V . Finally, the conclusions are presented in Sec. VI . II Problem Configuration and Imaging Operator In this section, we describe the single channel SAR configuration. A circular radar trajectory is considered a practical choice, as shown in [ 16 ] and [ 17 ] . We assume that the radar moves along a circular trajectory of radius R above a region of interest (ROI). The center of the radar trajectory is located above the center of the ROI. Fig. 1: System configuration. In Fig 1 , \beta_{i} represents the elevation angle, where i is the slow time index. Here, \mathbf{r}_{i}^{a} is the position of the antenna phase center at azimuth \theta_{i} , and \mathbf{r} denotes a point on the xy -plane. \mathbf{r}=(x,y,0) (1) The radar transmits at uniformly spaced angles \theta_{i} . Let V_{t}(f_{l}) denote the Fourier transform of the transmitted signal, and V_{r}(\theta_{i},f_{l}) denote the Fourier transform of the received signal at an azimuthal angle \theta_{i} (i=1,...,N_{\text{p}} ), both at evenly spaced frequencies f_{l} ( l=1,...,N_{f} ). The normalized received signal data is represented as an N_{f}\times N_{\text{p}} matrix P(\theta_{i},f_{l})=\frac{V_{r}(\theta_{i},f_{l})}{V_{t}(f_{l})} (2) and referred as the range profile matrix. We make the standard approximation that the radar does not move a significant distance while the pulse is being transmitted and received (”stop and go” assumption). In a near-field scenario, the reflectivity distribution of the xy -plane, denoted g(\mathbf{r}) , can be evaluated using the discrete backprojection operator. Assuming planar and horizontal ROI, this operator is defined as: g(\mathbf{r})=\sum_{i=1}^{N_{\text{p}}}\sum_{l=1}^{N_{f}}A(\mathbf{r},\mathbf{r}_{i}^{\text{a}},f_{l})^{-1}P(\theta_{i},f_{l})e^{j2k_{l}\mathinner{\!\left\lvert\mathbf{r}-\mathbf{r}_{i}^{\text{a}}\right\rvert}} (3) where k_{l}=2\pi f_{l}/c is the wave number with c being the speed of light. Also in ( 3 ), a slowly varying factor A is defined as A(\mathbf{r},\mathbf{r}_{i}^{\text{a}},f_{l})=\frac{F^{2}(\bm{\phi},f_{l})}{\mathinner{\!\left\lvert\mathbf{r}-\mathbf{r}_{i}^{\text{a}}\right\rvert}^{2}} (4) where \mathinner{\!\left\lvert\mathbf{r}-\mathbf{r}_{i}^{\text{a}}\right\rvert}^{2} accounts for the two-way propagation loss. Also in ( 4 ), F represents the field pattern of the radar antenna, which depends on the frequency f_{l} and the observation direction \bm{\phi}=(\phi_{\text{el}},\phi_{\text{az}}) . This observation direction is determined by the radar position, \mathbf{r}_{i}^{\text{a}} , and the imaging point \mathbf{r} . From this point on, we assume N_{f}=N_{\text{p}}=N . The reflectivity distribution is evaluated on a discrete, evenly spaced N\times N grid of spatial locations \mathbf{r} . Brute-force computation of the reflectivity using the discrete form of the BP operator ( 3 ) is characterized by O(N^{4}) complexity. Equation ( 3 ) can be modified to allow imaging of targets moving with constant velocity \mathbf{v}=(v_{x},v_{y},0) , as shown in [ 15 ] : g(\mathbf{r},\mathbf{v})=\sum_{i=1}^{N}\sum_{l=1}^{N}A(\mathbf{r},\mathbf{r}_{i}^{\text{a}},f_{l})^{-1}P(\theta_{i},f_{l})e^{j2k_{l}\mathinner{\!\left\lvert(\mathbf{r}+\mathbf{v}t_{i})-\mathbf{r}_{i}^{\text{a}}\right\rvert}} (5) where t_{i} is the slow time corresponding to \theta_{i} . Directly evaluating ( 5 ) on an N^{4} -point spatial location and velocity grid ( \mathbf{r},\mathbf{v} ) would require staggering O(N^{6}) operations. To that end, the following section introduces algorithms designed to reduce the CC. III Fast Algorithms III-A 4-D Multi-Level Imaging Algorithm A recursive multi-level algorithm, originally introduced in [ 9 ] for the static case, was adapted to the dynamic case through appropriate modifications described in [ 15 ] . This algorithm, known as the multi-level domain decomposition (MLDD) method, is further enhanced in this work by proposing an iterative implementation. In the initial phase, the input data is partitioned into subdomains of size N_{\text{c}}\times N_{\text{c}} . For each subdomain, a low-resolution image is computed using expression ( 5 ) on a discrete grid defined by (\mathbf{r}^{L_{1}},\mathbf{v}^{L_{1}})