Activity Classification Using Raw Range and I & Q Radar Data with Long Short Term Memory Layers

Abstract

This paper presents the first initial results of using radar raw I & Q data and range profiles combined with Long Short Term Memory layers to classify human activities. Although tested only on simple classification problems, this is an innovative approach that enables to bypass the conventional usage of Doppler-time patterns (spectrograms) as inputs of the LSTM layers, and adopt instead sequences of range profiles or even raw complex data as inputs. A maximum 99.56% accuracy and a mean accuracy of 97.67% was achieved by treating the radar data as these time sequences, in an effective scheme using a deep learning approach that did not require the pre-processing of the radar data to generate spectrograms and treat them as images. The prediction time needed for a given input testing sample is also reported, showing a promising path for real-time implementation once the LSTM network is properly trained.

Arm Motion Classification Using Time-Series Analysis of the Spectrogram Frequency Envelopes

“…The LSTM is an alternative RNN architecture which can overcome this shortcoming. A detailed explanation of LSTM can be found in [25][26][27][28].…”

“…Time-series analysis appears in many application domains, including speech recognition, handwriting recognition, weather readings, and financial recordings [18][19][20]. We consider two common time-series recognition methods, namely, the NN-dynamic time warping (DTW) (NN classifier with the DTW distance) [21][22][23][24] method and the long short-term memory (LSTM) method [25][26][27][28]. The former is a conventional machine learning (ML) technique that utilizes the DTW distance which is a sum-measure over a parametrization.…”

Radar for assisted living in the context of Internet of Things for Health and beyond

“…[69] 96.7 Decision level fuzzy logic [66] 94.8 Decision level voting [12] 97.8 Figure 3: sensitivity and specificity for the fall action using various classifications methods [12] Figure 3 shows an improvement overall without affecting fall specificity (reaching 100% with voting) by applying suitable feature selection and fusion. [70] LSTM are used to classify directly from raw data and range maps for binary classification every 0.5 s of action recorded. 5 subjects contributed, actions were recorded continuously for 60s giving 19 recordings (10 'walk' and 9 'sitting & standing').…”

“…right) illustrate I & Q data for walking movement. The time patterns that exist in the signals are exploited through the LSTM layers [70].…”

Human Activities Classification in a Complex Space Using Raw Radar Data

“…In this paper, we propose using Long Short Term Memory (LSTM) units in a recurrent neural network to classify six different activities, expanding from the preliminary results of simpler binary classification in our previous work [17]. As shown in Figure 1, 60s of radar data for six activities performed by six participants have been segmented as 0.5-second subsamples.…”