Joint Multitarget Detection and Tracking with mmWave Radar
Jiang Zhu, Menghuai Xu, Ruohai Guo, Fangyong Wang, Guangying Zheng, Fengzhong Qu · 2024
Read Joint Multitarget Detection and Tracking with mmWave Radar by Jiang Zhu, Menghuai Xu, Ruohai Guo, Fangyong Wang, Guangying Zheng, Fengzhong Qu (2024) 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-2412-17211.txt.
Introduction
Jiang Zhu and five colleagues build an end-to-end system that spots and tracks several targets with millimeter-wave radar. Such radar is a core sense for smart machines and automated spaces. Weak targets drown easily next to strong ones. Detection, matching, and tracking must work as one chain.
Their framework opens with a fast, sharp estimator that pulls out each target's distance and speed. A classic beam step then reads off the angles. A message-passing module matches returns to known tracks, using speed as a clue. A Kalman filter carries each track forward, guided by the estimator's known error shape. An extrapolation trick lifts the odds of catching faint targets. Tests on simulated and real data beat rival pipelines. The authors offer the chain as a practical core for radar perception.
Catalog record
| CREATOR | Jiang Zhu, Menghuai Xu, Ruohai Guo, Fangyong Wang, Guangying Zheng, Fengzhong Qu |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 78 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-17211
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Joint Multitarget Detection and Tracking with mmWave Radar Jiang Zhu, Menghuai Xu, Ruohai Guo, Fangyong Wang, Guangying Zheng, Fengzhong Qu Abstract Accurate targets detection and tracking with mmWave radar is a key sensing capability that will enable more intelligent systems, create smart, efficient, automated system. This paper proposes an end-to-end detection-estimation-track framework named MNOMP-SPA-KF consisting of the target detection and estimation module, the data association (DA) module and the target tracking module. In the target estimation and detection module, a low complexity, super-resolution and constant false alarm rate (CFAR) based two dimensional multisnapshot Newtonalized orthogonal matching pursuit (2D-MNOMP) is designed to extract the multitarget’s radial distances and velocities, followed by the conventional (Bartlett) beamformer to extract the multitarget’s azimuths. In the DA module, a sum product algorithm (SPA) is adopted to obtain the association probabilities of the existed targets and measurements by incorporating the radial velocity information. The Kalman filter (KF) is implemented to perform target tracking in the target tracking module by exploiting the asymptotic distribution of the estimators. To improve the detection probability of the weak targets, extrapolation is also coupled into the MNOMP-SPA-KF. Numerical and real data experiments demonstrate the effectiveness of the MNOMP-SPA-KF algorithm, compared to other benchmark algorithms. I Introduction Target detection and tracking play pivotal roles across various domains, including computer vision, robotics, military and defense, surveillance and security, and autonomous systems. Undoubtedly, the efficacy of target detection and tracking applications is based on the paramount considerations of accuracy and precision. Although sensors like passive infrared (PIR) and time of flight (TOF) bring their unique advantages, they are not without limitations that can impact their performance. For example, PIR sensors, adept at detecting infrared radiation emitted by objects, offer a cost-effective and straightforward implementation. However, their measurements are sensitive to temperature fluctuations, rendering them susceptible to false alarms triggered by environmental changes such as shifts in sunlight and heat sources. However, TOF sensors deliver highly accurate distance information but struggle with challenges such as false alarms due to interference from reflective surfaces and susceptibility to environmental changes such as darkness, smoke, or fog [ 1 ] . Undoubtedly, millimeter-wave (mmWave) radar technology offers numerous advantages, rendering it highly suitable for diverse applications. MmWave radar excels at providing precise measurements of the position and velocity of objects, including individuals, within a specified area. In contrast to PIR or TOF sensors, which may be influenced by environmental conditions such as rain, dust, and smoke, mmWave radar maintains optimal performance even in adverse weather conditions. This reliability extends to both outdoor and indoor environments. Furthermore, mmWave radar sensors exhibit the capability to operate seamlessly in complete darkness or bright daylight. This versatility makes mmWavee radar an ideal solution for applications that require uninterrupted monitoring, regardless of the time of day [ 2 ] . Although mmWave radar has unique advantages over PIR or TOF sensors in target sensing and tracking, it encounters challenges, particularly amidst strong background clutters. In such scenarios, the reflected signal from weaker targets, such as pedestrians with a small radar cross-section (RCS), often gets overshadowed by the reflected waves from formidable targets like surrounding buildings and trucks. Consequently, these weaker targets may go unnoticed. Additionally, the widely recognized issue lies in the estimation and detection algorithm, which tends to generate numerous false alarms, imposing a substantial computational burden to manage and alleviate their impact on overall system performance. To tackle the first challenge, a low-complexity, super-resolution compressed sensing (CS) based algorithm incorporating constant false alarm rate (CFAR) should be developed. This addresses target-masking effects, enabling the estimation and detection of weak targets near strong targets. Addressing the second challenge involves the integration of the detection, estimation, and tracking algorithms. This integration ensures that false alarms generated in the detection stage are significantly suppressed during the tracking stage, by allowing a higher false alarm probability in the detection stage. Both serve as the motivation behind our work. I-A Related Works The joint target detection, estimation, and tracking algorithm encompasses the extraction of targets’ states, comprising azimuth, radial distance, and radial velocity, from the baseband data. This algorithm proceeds by conducting target detection to ascertain the number of targets, executing data association among the targets’ states across various time instants, and employing the Kalman filter (KF) for single target tracking. Given the ubiquity of the KF as a standard algorithm, our focus lies predominantly on detailing the relevant works pertaining to target estimation and detection, as well as the data association (DA) algorithm. Additionally, Texas Instruments (TI) has presented a system solution for people tracking and counting utilizing mmWave radar sensors [ 3 , 4 ] , and this solution will also be elaborated upon in this discussion. I-A 1 Targets Detection and Estimation The linear frequency modulated continuous wave (LFMCW) mmWave radar operates by transmitting a chirp signal, and its receiver, equipped with a uniform linear array (ULA), captures the signal reflected by targets within the field of view. Following the dechirping process, the received signal can be conceptualized as a three-dimensional line spectral. Extracting frequencies from this line spectral allows for the estimation of the target’s states. Consequently, our focus in this context is on reviewing the pertinent literature regarding line spectral estimation (LSE) and detection. The fast Fourier transform (FFT) algorithm [ 5 ] and the constant false alarm rate (CFAR) detector [ 6 ] are widely recognized as the most commonly used method for extracting frequencies and target detection, respectively. Despite their simplicity and effectiveness, they do come with limitations. The estimation accuracy of FFT is constrained by the grid spacing, limiting its performance [ 7 ] . In addition, the CFAR detector faces challenges where a weak target near a strong target might be missed due to the strong sidelobe of the latter. To overcome these issues, compressive sensing (CS) methods have been introduced. These encompass variational line spectra estimation (VALSE) [ 8 ] , atomic norm soft thresholding (AST) [ 9 ] , Newtonized orthogonal matching pursuit (NOMP) [ 10 ] , CFAR-based NOMP (NOMP-CFAR) [ 11 ] , among others. These methods leverage the sparse nature of targets in the range, Doppler, and spatial domains. Among them, NOMP stands out for its speed but necessitates knowledge of the noise variance. NOMP-CFAR integrates an adaptive CFAR detector into NOMP, eliminating the need for noise variance information but at the cost of increased computational complexity due to forward-backward steps. In contrast, this paper proposes a low-complexity NOMP-CFAR based scheme that is fast and does not require knowledge of the noise variance. I-A 2 Data Association The challenge of data association and target tracking revolves around extracting target trajectories based on sequences of sensor measurements [ 12 ] . The nearest neighbors (NN) algorithm [ 13 ] , a classical tracking approach, achieves target tracking by matching the current measurement closest to the predicted measurement generated by the target. While it boasts speed and simplicity, it overlooks uncertainties in data association, performing suboptimally in low SNR or cluttered scenarios. The probabilistic data association (PDA) algorithm [ 14 ] , calculating the average of measurements weighted by association probability, also offers speed and simplicity and excels in scenarios with a single target.