Self-similarity matrix based slow-time feature extraction for human target in high-resolution radar

Abstract

Self-similarity matrix based slow-time feature extraction for human target in high-resolution radar yuan he, pascal aubry, francois le chevalier and alexander yarovoy A new approach is proposed to extract the slow-time feature of human motion in high-resolution radars. The approach is based on the self-similarity matrix (SSM) of the radar signals. The Mutual Information is used as a measure of similarity. The SSMs of different radar signals (high-resolution range profile, micro-Doppler, and range-Doppler video sequence) are compared, and the angel-invariant property of the SSMs is demonstrated. The SSM for different activities (i.e. walking and running) is extracted from range-Doppler video sequence and analyzed. Finally, simulation result is validated by experimental data.Keywords: Self-similarity matrix, Slow-time feature, Human target analysis, High-resolution radar . As a tool to illustrate Doppler spectra along the slow-time axis, the microDoppler images have been studied widely due to their potential in human classification [5]. The range-Doppler (RD) images were also used to analyze distributed human scatterers [2]. Although HHRP, micro-Doppler, and RD images all show certain target information, they are restricted to observe targets in either range, Doppler or RD, and the target slowtime evolution has not been addressed in all these signals.Slow-time behavior for human motions (e.g. walking, crawling, and running) is unique. For example, the standard walking procedure can normally be seen as a periodic movement with a certain cadence frequency. Human periodic motion has been recently analyzed in camera-based systems [6,7]. Cutler and Davis [6] proposed using self-similarity matrix (SSM) of one optical image sequence to detect and analyze periodic motions. For action recognition, Junejo et al. [8] claimed that an important structural stability of SSM can be found for a moving person observed by different cameras. Although radar signals differ significantly from optical images, in this paper we will demonstrate that the SSM theory can be extended for radar signals with some adequate modifications.Our approach is to apply SSM to extract the slow-time feature of typical human radar backscattering (i.e. HRRP, microDoppler image, and RD video sequence). Range-Doppler video sequence (RDVS) [9] is one sequence of RD images, as a function of slow-time. RDVS not only preserves the target range information, but also keeps the Doppler information, as a function of slow-time. The SSMs obtained from different signals will be analyzed, and their angle-invariant characteristic will be demonstrated by comparing the SSMs from radars deployed at different locations. Fourier transform-based periodicity detection method will be developed to extract the cadence frequency of the human gait from SSMs. Finally, the proposed approach will be validated on experimental data. This paper is organized as follows. The radar backscattering from the walking human model is described in Section II. Section III discusses SSM of human backsc...

Efficient hierarchical temporal segmentation method for facial expressionsequences

“…Temporal video segmentation aims at splitting a video sequence into homogeneous subsequences to make the properties of each subsequence different. It can be considered as a clustering problem, which has been widely used in human behavior segmentation [6][7][8][9][10][11][12] and face clustering [5,[13][14][15][16][17][18]. One important condition expected of a clustering algorithm is that the intracluster variance should be low.…”

“…The SSM, which is a variant of the recurrence plots (RPs) [12], is used widely to describe the texture features between sequences. Junejo et al [8] explored self-similarities of action sequences over time to recognize human behaviors in different perspectives.…”

Range‐Doppler surface: a tool to analyse human target in ultra‐wideband radar

“…Defined as a sequence of RD images, the RD video sequence [12] was proposed to describe the slow-time evolution of target RD signatures. One pulse repetition interval (PRI) can be used as a typical interval between two frames.…”

“…To simplify the derivation, replacing the discrete slow time mT r with t s in (12) and denoting 2v/c as β, the reflected signal in fast-time/ slow-time domain is defined as…”

Toward Deep Learning-Based Human Target Analysis

“…Therefore, in order to analyze the target signature more comprehensively, we describe a new concept called range-Doppler surface (RDS). As an alternative to the micro-Doppler profile and HRRP, RDS is a radar target representation extracted from a three-dimensional data cubethe range-Doppler (RD) video sequence [11,12]. The RDS consists of all the important information contained in both HRRP and micro-Doppler signatures.…”