Baris Erol

A kinect-based human micro-doppler simulator

Until recently, human surveillance has primarily been accomplished using video cameras. However, radar offers unique advantages over optical sensors, such as being able to operate at far distances, under adverse weather conditions, and at nighttime, when optical devices are unable to acquire meaningful data. Radar is capable of recognizing human activities by classifying the micro-Doppler signature of a subject. Micro-Doppler is caused by any rotating or vibrating parts of a target, and results in frequency modulations centered about the main Doppler shift caused by the translational motion of the target [1]. Thus, the rotation of a helicopter blade, wheels of a vehicle, or treads of a tank all result in micro-Doppler. In the case of humans, the complex motion of the limbs that occur in the course of any activity all result in a micro-Doppler signature visually distinguishable from other targets, even animals [2]-[3], which can then be exploited for human detection [4]-[5], automatic target recognition (ATR) [6]-[7], and activity classification [8].

Range information for reducing fall false alarms in assisted living

Fall detection using radar technology is emerging to be an area of significance research and development. The main challenges remain in reducing false alarms to a level that is acceptable to healthcare providers and retirement homes. One source of false alarms is the confusion of actual falls and fall-like gross-motor activities, such as sitting. Discriminative features based on only Doppler frequency may declare fast sitting as a slow fall. However, sittings, unlike falls, have limited range extent which can be easily revealed by using a range-Doppler radar, in lieu of continuous wave (CW) radar. Using experimental data from two different ultra-wideband (UWB) radars operating different carrier frequencies as well as data from Kinect-based UWB pulse-Doppler radar simulator, this paper demonstrates reductions in fall false alarm rates exploiting the range information compared to Doppler only processing.

Radar fall detectors: a comparison

Falls are a major cause of accidents in elderly people. Even simple falls can lead to severe injuries, and sometimes result in death. Doppler fall detection has drawn much attention in recent years. Micro-Doppler signatures play an important role for the Doppler-based radar systems. Numerous studies have demonstrated the offerings of micro-Doppler characteristics for fall detection. In this respect, a plethora of micro-Doppler signature features have been proposed, including those stemming from speech recognition and wavelet decomposition. In this work, we consider four different sets of features for fall detection. These can be categorized as spectrogram based features, wavelet based features, mel-frequency cepstrum coefficients, and power burst curve features. Support vector machine is employed as the classifier. Performance of the respective fall detectors is investigated using real data obtained with the same radar operating resources and under identical sensing conditions. For the considered data, the spectrogram based feature set is shown to provide superior fall detection performance.

Fall motion detection using combined range and Doppler features

Feature selection based on combined Doppler and range information improves fall detection and enables better discrimination against similar high Doppler non-rhythmic motions, such as sitting. A fall is typically characterized by an extension in range beyond that associated with sitting, which is determined by the seat horizontal depth. In this paper, we demonstrate, using time-frequency (TF) spectrograms, that range-Doppler radar plays a fundamental and important role in motion classification for assisted living applications. It reduces false alarms along with the associated cost in the unnecessary deployment of the first responders. This reduction is considered vital for the development of in-home radar monitoring and for casting it as a viable technology for aging-in-place.

Wideband radar based fall motion detection for a generic elderly

Radar-based automated fall detection systems are considered as an important and emerging technology for elderly assisted living. These radar systems provide non-intrusive sensing capabilities to detect fall events. Various studies have used micro-Doppler signatures to determine falls. However, Doppler radar fall detection systems suffer false alarms stemming from other sudden non-rhythmic motion articulations. In this work, we consider a textural-based feature extraction method which can determine the density variations between various motion articulations. For this purpose, textural features are extracted from the gray level co-occurrence matrix for each motion using time-integrated range-Doppler maps and micro-Doppler signatures. Textural features are then used to train the support vector machine classifier. The sequential forward selection method is implemented to identify essential features and minimize the feature space while maximizing the fall detection rate. The results show that well selected range-Doppler based textural features can provide improved classification results compared to textural features based only on micro-Doppler signatures.

Multiple joint-variable domains recognition of human motion

Radar has been successfully employed for classifying human motions in defense, security and civilian applications, and has emerged to potentially become a technology of choice in the healthcare industry, specifically in what pertains to assisted living. Due to the relationship between Doppler frequency and motion kinematics, the time-frequency domain has been traditionally used to analyze radar signals of human gross-motor activities. Towards improving motion classification, this paper incorporates three domains, namely, time-frequency, time-range, and range-Doppler domains. Features from each domain are extracted using deep neural network that is based on stacked auto-encoders. Final decision is made by combining the classification outcomes. Experimental results demonstrate that certain domains are more favorable than others in recognizing specific motion articulations, thus reinforcing the merits of multi-domain motion classifications.

Range-Doppler radar sensor fusion for fall detection

Falls are the major cause of accidents in the elderly population. Propelled by their non-intrusive sensing capabilities and robustness to heat and lighting conditions, radar-based automated fall detection systems have emerged as a candidate technology for reliable fall detection in assisted living. The use of a multiple radar system, in lieu of a single radar unit, for indoor monitoring combats occlusion and supported by the fact that motion articulations in the directions away from the line of sight generate weak Doppler signatures that are difficult to detect and classify. Fusion of the data from two radars is deemed to improve performance and reduce false alarms. Utilizing two 24 GHz ultra-wide band (UWB) radar sensing systems, we present different fusion architectures and sensor selection methods, demonstrating the merits of two-sensor platform for indoor motion monitoring and elderly care applications.

Effects of range spread and aspect angle on radar fall detection

Single-sensor Doppler radar faces many challenges in elderly fall detection. The similarities of the Doppler signatures between falls and other motions of fast time transitions render CW-based EM sensing insufficient for proper motion discriminator. Further, motion articulations in the directions of a small or zero projections along the line-of-sight generate noisy Doppler signatures and greatly contribute to the confusion matrix. In this paper, the benefit from the range information is used to distinguish between fall and non-fall motions with similar Doppler features. We also examine the effects of the aspect angle on fall detection. Simulation results using Kinect-based radar simulator and 24 GHz UWB radar sensing system, demonstrate the merits of the proposed platform for indoor motion monitoring serving assisted living applications.

Data-driven cepstral and neural learning of features for robust micro-Doppler classification

Automatic target recognition (ATR) using micro-Doppler analysis is a technique that has been a topic of great research over the past decade, with key applications to border control and security, perimeter defense, and force protection. Patterns in the movements of animals, humans, and drones can all be accomplished through classification of the target’s micro-Doppler signature. Typically, classification is based on a set of fixed, pre-defined features extracted from the signature; however, such features can perform poorly under low signal-to-noise ratio (SNR), or when the number and similarity of classes increases. This paper proposes a novel set of data-driven frequency-warped cepstral coefficients (FWCC) for classification of micro-Doppler signatures, and compares performance with that attained from the data-driven features learned in deep neural networks (DNNs). FWCC features are computed by first filtering the discrete Fourier Transform (DFT) of the input signal using a frequency-warped filter bank, and then computing the discrete cosine transform (DCT) of the logarithm. The filter bank is optimized for radar using genetic algorithms (GA) to adjust the spacing, weight, and width of individual filters. For a 11-class case of human activity recognition, it is shown that the proposed data-driven FWCC features yield similar classification accuracy to that of DNNs, and thus provides interesting insights on the benefits of learned features.

Data-dependent micro-Doppler feature selection

A vast number of features have been proposed over the years for classification of radar micro-Doppler signatures. However, the degree to which a feature may contribute in discriminating between classes depends upon a variety of operational considerations, such as antenna-target aspect angle, signal-to-noise ratio (SNR), and dwell time. Moreover, utilization of all features in every circumstance does not necessarily ensure optimal classification performance. Oftentimes a well-selected subset of robust features yield better results. In this work, the variance of micro-Doppler feature estimates are examined under a variety of operational conditions and used to select feature subsets. The classification performance of data-dependent feature subsets are compared to that attained without any feature selection. Results show that data-dependent feature selection yields higher correct classification rates over a wider range of operational situations.