Accurate and automated Range of Motion (ROM) assessment is essential for rehabilitation, physical therapy, and post-surgical recovery. Traditional manual goniometer-based evaluations suffer from subjectivity, inter-rater variability, and reliance on trained professionals. Although RGB-based computer vision enables markerless ROM estimation, it remains susceptible to occlusions, pose estimation errors, and difficulties detecting subtle joint movements. Furthermore, vision alone cannot capture neuromuscular activation, which is crucial for understanding joint dynamics. To address these challenges, we propose a multi-modal deep learning framework that integrates RGB-based motion tracking with electromyography (EMG) signals. EMG provides neuromuscular activation data, enhancing the robustness against visual occlusions and improving sensitivity to subtle joint displacements. Our method employs an Hourglass-based convolutional neural network (CNN) for spatial feature extraction and a gated recurrent unit (GRU)-based model for temporal EMG processing. To further enhance performance, we introduce feature-level and modality-level attention modules, dynamically emphasizing the most informative features and modality contributions. Experimental results demonstrate that our proposed model achieves an overall RMSE of 2.55, and improvements via the feature and modality attention mechanisms, respectively. Moreover, the fully fused RGB-EMG model outperforms RGB-only approaches, particularly in accurately predicting subtle ROM movements.

The increasing connectivity of modern vehicles exposes them to growing cybersecurity threats, driving the need for robust and efficient in-vehicle intrusion detection systems (IDSs). Due to the time-critical nature of in-vehicle communication, real-time responsiveness is a key requirement for in-vehicle IDSs. While lightweight deep learning-based IDS models are suitable for real-time deployment, they typically require manual design and tuning - a process that is time-consuming and often suboptimal. Recent advances in neural architecture search (NAS) offer a promising alternative by automating model discovery. However, most NAS approaches overlook hardware-specific latency constraints, resulting in architectures that perform well only in theory. To address this, we propose a hardware-aware NAS (HW-NAS) framework that incorporates real latency measurements from target embedded platforms (Raspberry Pi 4 and 5, serving as ECU proxies) via latency lookup tables (LUTs). This enables joint optimization of detection accuracy and inference efficiency during search. Experimental results on two CAN bus intrusion datasets show that the proposed HW-NAS model achieves up to 99.63% accuracy with a 0.99 F1 score, while delivering the lowest latency (2.8 ms) and smallest parameter count (27K) among all evaluated models. These results demonstrate the framework's effectiveness in producing real-time, resource-efficient IDS models tailored for embedded automotive platforms.

Modern vehicles are equipped with multiple Electronic Control Units (ECUs) communicating over in-vehicle networks such as Controller Area Network (CAN). Inherent security limitations in CAN necessitate the use of Intrusion Detection Systems (IDSs) for protection against potential threats. While some IDSs leverage advanced deep learning to improve accuracy, issues such as long processing time and large memory size remain. Existing Binarized Neural Network (BNN)-based IDSs, proposed as a solution for efficiency, often compromise on accuracy. To this end, we introduce a novel Binarized Convolutional Neural Network (BCNN)-based IDS, designed to exploit the temporal and spatial characteristics of CAN messages to achieve both efficiency and detection accuracy. In particular, our approach includes a novel input generator capturing temporal and spatial correlations of messages, aiding model learning and ensuring high-accuracy performance. Experimental results suggest our IDS effectively reduces memory utilization and detection latency while maintaining high detection rates. Our IDS runs 4 times faster and utilizes only 3.3% of the memory space required by a full-precision CNN-based IDS. Meanwhile, our proposed system demonstrates a detection accuracy between 94.19% and 96.82% relative to the CNN-based IDS across different attack scenarios. This performance marks a noteworthy improvement over existing state-of-the-art BNN-based IDS designs.

Upper limb kinematic analysis that has been employed in the clinical assessment of motion functions or rehabilitation training is traditionally tested manually with a goniometer. Nowadays, it is a trend to deploy different technology and devices including low-cost but accurate RGB cameras in order to save manual efforts. Among these, a new method using deep learning-based cameras has been investigated to provide the same ease and accessibility as a manual handheld goniometer. The key to measuring upper limb Range of Motion (ROM) using a camera is to estimate upper limb joints accurately. Many existing joint estimation algorithms focus on improving the accuracy performance but put the efficiency concerns aside. It is still challenging to apply those algorithms to low-capacity and budget-friendly devices, which is highly demanding in clinical scenarios. We propose a lightweight and fast deep learning model to estimate human pose and then use predicted joints to measure the range of motion for upper limb joints. Unlike other human pose estimation methods that learn and predict all major joints of the human body, the proposed model only focuses on the upper limb, which improves the accuracy and reduces the overhead of prediction. To further reduce model size and latency, our model is based on a compact neural network architecture, and parameters in the network are quantized to 8-bit precision. As a result, our model runs 4.1 times faster and is 15.5 times smaller compared with a full sized state of the art human pose estimation model. The proposed method is further evaluated on different upper limb functional tasks. Results show that our new method achieves a satisfying accuracy in ROM measurement and a high degree of agreement with a goniometer. Compared with the goniometer to measure ROM, our presented method is easier to operate and can be performed remotely, while still retaining good accuracy.