Maintaining a safe distance from the vehicle ahead is critical for safe driving. While LiDAR sensors can be used for distance measurement, their high cost and the large number of vehicles without such sensors—especially in developing regions—necessitate a low-cost yet effective alternative. To address this, we introduce CarVision, a system that utilizes a commercially available single-chip mmWave radar mounted on a car’s dashboard for vehicle ranging. CarVision detects vehicles within the radar’s field of view (FoV) and computes both the distance and relative speed of the vehicles in front. Our system employs a novel hybrid vehicle detection and tracking approach, combining deep neural network-based detection with range tracking in a streamlined pipeline to optimize accuracy and speed. Additionally, we’ve developed a smartphone-based alert system that warns drivers if a vehicle approaches within a critical distance (approximately 2 meters). CarVision demonstrates reliable ranging performance in daytime and nighttime conditions, with accurate measurements up to 50 meters.

In this work, we introduce RadarTrack, an innovative ego-speed estimation framework utilizing a single-chip millimeter-wave (mmWave) radar to deliver robust speed estimation for mobile platforms. Unlike previous methods that depend on cross-modal learning and computationally intensive Deep Neural Networks (DNNs), RadarTrack utilizes a novel phase-based speed estimation approach. This method effectively overcomes the limitations of conventional ego-speed estimation approaches which rely on doppler measurements and static surroundings. RadarTrack is designed for low-latency operation on embedded platforms, making it suitable for real-time applications where speed and efficiency are critical. Our key contributions include the introduction of a novel phase-based speed estimation technique solely based on signal processing and the implementation of a real-time prototype validated through extensive real-world evaluations. By providing a reliable and lightweight solution for ego-speed estimation, RadarTrack holds significant potential for a wide range of applications, including micro-robotics, augmented reality, and autonomous navigation.

In this demonstration we introduce a novel approach to ego-speed estimation using a single-chip Commercial-Off-the-Shelf (COTS) millimeter-wave (mmWave) radar. Contrary to previous approaches that are cross-modal learning based and dependent on the computationally expensive nature of DNN's, our proposed approach RadarTrack is based on a phase-based approach to speed estimation. Our approach successfully overcomes the drawbacks of traditional ego-speed estimation methods that are doppler-based and static environment dependent. RadarTrack is intended to support low-latency execution on embedded systems such that it can be used in real-time applications where efficiency and speed are equally important. We have also created a real-time visualizer which is capable of recording the phasebased ego-speed together with state-of-the-art doppler-based speed estimation and comparatively demonstrate how phasebased ego-speed estimation can better record the speed of an ego-vehicle.

Continuous detection of human activities and presence is essential for developing a pervasive interactive smart space. Existing literature lacks robust wireless sensing mechanisms capable of continuously monitoring multiple users’ activities without prior knowledge of the environment. Developing such a mechanism requires simultaneous localization and tracking of multiple subjects. In addition, it requires identifying their activities at various scales, some being macro-scale activities like walking, squats, etc., while others are micro-scale activities like typing or sitting, etc. In this paper, we develop a holistic system called MARS using a single Commercial off-the-shelf (COTS) Millimeter Wave (mmWave) radar, which employs an intelligent model to sense both macro and micro activities. In addition, it uses a dynamic spatial time-sharing approach to sense different subjects simultaneously. A thorough evaluation of MARS shows that it can infer activities continuously with an accuracy of > 93% and an average response time of ≈ 2 sec, with 5 subjects and 19 different activities.

Developing robust wireless sensing mechanisms for continuously monitoring human activities and presence is crucial for creating pervasive interactive intelligent spaces. The existing literature lacks solutions that continuously monitor multiple users’ activities without prior knowledge of the environment. This requires simultaneous localization and tracking of multiple subjects and identifying their activities at various scales, including macro-scale activities like walking and squats and micro-scale activities like typing or sitting. In this demo, we present MARS , a holistic system using a single off-the-shelf mmWave radar. MARS employs an intelligent model to sense both macro and micro activities and uses a dynamic spatial time-sharing approach to sense different subjects simultaneously. Our thorough evaluation demonstrates that MARS can continuously infer activities with over 93% accuracy and an average response time of approximately 2 seconds, even with five subjects performing 19 different activities.

Precise ego-motion measurement is crucial for various applications, including robotics, augmented reality, and autonomous navigation. In this poster, we propose mmPhase, an odometry framework based on single-chip millimetre-wave (mmWave) radar for robust ego-motion estimation in mobile platforms without requiring additional modalities like the visual, wheel, or inertial odometry. mmPhase leverages a phase-based velocity estimation approach to overcome the limitations of conventional doppler resolution. For real-world evaluations of mmPhase we have developed an ego-vehicle prototype. Compared to the state-of-the-art baselines, mmPhase shows superior performance in ego-velocity estimation.

Continuous tracking of eye movement dynamics plays a significant role in developing a broad spectrum of human-centered applications, such as cognitive skills (visual attention and working memory) modeling, human-machine interaction, biometric user authentication, and foveated rendering. Recently neuromorphic cameras have garnered significant interest in the eye-tracking research community, owing to their sub-microsecond latency in capturing intensity changes resulting from eye movements. Nevertheless, the existing approaches for event-based eye tracking suffer from several limitations: dependence on RGB frames, label sparsity, and training on datasets collected in controlled lab environments that do not adequately reflect real-world scenarios. To address these limitations, in this paper, we propose a dynamic graph-based approach that uses a neuromorphic event stream captured by Dynamic Vision Sensors (DVS) for high-fidelity tracking of pupillary movement. More specifically, first, we present EyeGraph, a large-scale multi-modal near-eye tracking dataset collected using a wearable event camera attached to a head-mounted device from 40 participants -- the dataset was curated while mimicking in-the-wild settings, accounting for varying mobility and ambient lighting conditions. Subsequently, to address the issue of label sparsity, we adopt an unsupervised topology-aware approach as a benchmark. To be specific, (a) we first construct a dynamic graph using Gaussian Mixture Models (GMM), resulting in a uniform and detailed representation of eye morphology features, facilitating accurate modeling of pupil and iris. Then (b) apply a novel topologically guided modularity-aware graph clustering approach to precisely track the movement of the pupil and address the label sparsity in event-based eye tracking. We show that our unsupervised approach has comparable performance against the supervised approaches while consistently outperforming the conventional clustering approaches.

Detecting dangerous driving has been of critical interest for the past few years. However, a practical yet minimally intrusive solution remains challenging as existing technologies heavily rely on visual features or physical proximity. With this motivation, we explore the feasibility of purely using mm Wave radars to detect dangerous driving behaviors. We first study characteristics of dangerous driving and find some unique patterns of range-doppler caused by 9 typical dangerous driving actions. We then develop a novel Fused-CNN model to detect dangerous driving instances from regular driving and classify 9 different dangerous driving actions. Through extensive experiments with 5 volunteer drivers in real driving environments, we observe that our system can distinguish dangerous driving actions with an average accuracy of 97(±2)%. We also compare our approach with existing state-of-the-art baselines to establish their significance,

Adaptive Bitrate (ABR) Streaming over cellular networks has been well studied in the literature; however, existing ABR algorithms primarily focus on improving the end-user Quality of Experience (QoE) while ignoring the resource consumption aspect of the underlying device. Consequently, proactive attempts to download video data to maintain the user’s QoE often impact the battery life of the underlying device unless the download attempts are synchronized with the network’s channel condition. In this work, we develop EnDASH-5G – a wrapper over the popular DASH-based ABR streaming algorithm, which establishes this synchronization by utilizing a network-aware video data download mechanism. EnDASH-5G utilizes a novel throughput prediction mechanism for 5G mmWave networks by upgrading the existing throughput prediction models with a transfer learning-based approach, leveraging publicly available 5G datasets. It then exploits deep reinforcement learning to dynamically decide the playback buffer length and the video bitrate using the predicted throughput. This ensures that the data download attempts get synchronized with the underlying network condition, thus saving the device’s battery power. From a thorough evaluation of EnDASH-5G, we observe that it achieves a near 30.5% decrease in the maximum energy consumption than the state-of-the-art Pensieve ABR algorithm while performing almost at par in term of QoE.