Widely used in automotive safety, industrial automation, and surveillance, Frequency-Modulated Continuous-Wave (FMCW) radar provides accurate object tracking (OT). Kalman filter (KF)-based algorithms are proven to work reliably in industry for OT with FMCW radar. Extending the KF functionality, Probabilistic Data Association Filter (PDAF) is a method that also enables the association of detected reflections with the observed target. Integrated PDAF (IPDAF) offers formulations for both the probability of track existence and data association simultaneously. This paper shows how parameters for IPDAF tracker can be selected to optimize the First, FMCW radar is simulated to generate precise distance, velocity, and angle measurements through a multiple-input multiple-output (MIMO) antenna setup providing ground-truth labeled data. Next, a practical method for calculating optimal tracking parameters (OTP), particularly for short-range applications, based on Optuna optimizer is introduced. This method aims to improve tracking accuracy in scenarios such as indoor environments and pedestrian safety systems, considering scenarios with different target movement maneuvers and different dynamic movement models.
This paper investigates two approaches for real-time object detection and classification using Frequency-Modulated Continuous-Wave (FMCW) radar data from the Radical Dataset. The first method uses the Constant False Alarm Rate (CFAR) algorithm and manual annotations to extract patches from Range-Azimuth (RA) heatmaps, which are then processed by a simple Convolutional Neural Network (CNN) named RADICAL-CNN. The second approach evaluates the performance of the RAMP-CNN model on the Radical Dataset, which incorporates RA, Range-Doppler (RD) and Doppler-Azimuth (DA) data and utilizes 3D convolution to integrate temporal information. This comparative analysis of two innovative neural network architectures not only highlights their strengths and weaknesses, but also contributes to the ongoing optimization of object detection techniques for FMCW radar data in dynamic, real-world scenarios such as autonomous driving, robotics, zone monitoring and safety.
Several research papers showed that long short-term memory (LSTM) recurrent neural networks (RNNs) are a powerful tool in algorithmic composing. They can learn both fine (short-term) and coarse (long-term) structures in music. Most of the promising results were got by using musical notation as input data. Can LSTMs give similar results by using time-series data, such as audio? This research performs a baseline experiment of an LSTM composing algorithm using audio data as input instead of musical notation. The audio data is presented to the LSTM as a chroma feature encoded to an integer representation. The LSTM is trained on a single piano piece with a segment held out. After, it is tasked with generating a song of a desired duration based on the held out segment. The experimental results show the LSTM has learned the coarse structure of the audio data (the key), but has not learned the fine structure (the melody). Because of this and degrading the audio quality when reconstructing from chroma features, the resulting output does not sound like music. The authors propose the results show only a proof of concept for the algorithm and that more work is needed.