Wearable devices for ECG arrhythmia detection based on artificial neural networks (ANN) are very popular. However, the energy consumption of electrocardiogram (ECG) processing in ANN has become one of the most critical factors. One solution is using a spiking neural network (SNN), effectively reducing power consumption and improving energy efficiency. Nevertheless, the inevitable membrane potential storage and accumulation of SNN result in significant energy and area overheads. This paper proposes a time domain (TD) based SNN processor for ECG classification. We propose a novel memory delay unit (MDU), part of the memory delay line (MDL), to store and accumulate membrane potential. With this method, power consumption can be significantly reduced. Also, we propose a wave generator that works with MDL to maximize computing efficiency. Compared with digital neurons, our proposed TD neurons reduce power consumption by 32.5% and achieve a classification accuracy of 96.8%. It is very suitable for arrhythmia detection wearable devices.

Loop Closure Detection (LCD) is an essential part of minimizing drift due to the accumulation of previously pose errors in Simultaneous Localization and Mapping (SLAM). The existing loop detection methods are limited by the changes of external conditions such as illumination, viewpoint and appearance. Previous work has mainly focused on the feature descriptor matching methods, which usually only consider the keypoints themselves. Here, we propose a fusion method GraSS, which uses the Graph Neural Network (GNN) based on visual features, and introduces semantics and depth, so as to enhance the spatial characteristics of the keypoints and the information correlation between them in the graph. Furthermore, a learnable parameter is added when two keypoints share the same semantic labels, their matching scores are increased, mitigating to some extent the issue of mismatch caused by significant differences in external conditions between two keypoints that should ideally be paired. Our findings show that GraSS has better performance than other state-of-the-art LCD methods when facing obvious illumination, appearance changes and slight viewpoint changes.

Computing-in-memory (CIM) architecture has become a possible solution to designing an energy-efficient artificial intelligent processor. Various CIM demonstrators indicated the computing efficiency of CIM macro and CIM-based processors. However, previous studies mainly focus on macro optimization and low CIM capacity without considering the weight update strategy of CIM architecture. The artificial intelligence (AI) processor with a CIM engine practically induces issues, including updating memory data and supporting different operators. For instance, AI-oriented applications usually contain various weight parameters. The weight stored in the CIM architecture should be reloaded for the considerable gap between the capacity of CIM and growing weight parameters. The computation efficiency of the CIM architecture is reduced by the weight updating and waiting. In addition, the natural parallelism of CIM leads to the mismatch of various convolution kernel sizes in different networks and layers, which reduces hardware utilization efficiency. In this work, we develop a CIM engine with a ping-pong computing strategy as an alternative to typical CIM macro and weight buffer, hiding the data update latency and improving the data reuse ratio. Based on the ping-pong engine, we propose a flexible CIM architecture adapting to different sizes of neural networks, namely, intelligent pong computing-in memory (IPOCIM), with a fine-grained data flow mapping strategy. Based on the evaluation, IPOCIM can achieve a 1.27–$6.27\times $performance and 2.34–$5.30\times $energy efficiency improvement compared to the state-of-the-art works.

Recently wearable EEG monitoring devices with seizure detection processor using convolutional neural network (CNN) have been proposed to detect the seizure onset of patients in real time for alert or stimulation purpose. High energy efficiency and accuracy are required for the seizure detection processor due to the tight energy constraint of wearable devices. However, the use of CNN and multi-channel processing nature of seizure detection result in significant energy consumption. In this work, an energy-efficient seizure detection processor is proposed, featuring multi-stage CNN classification, segmented data processing and adaptive channel selection to reduce the energy consumption while achieving high accuracy. The design has been fabricated and tested using a 55nm process technology. Compared with several state-of-the-art designs, the proposed design achieves the lowest energy per classification (0.32 μJ) with high sensitivity (97.78%) and low false positive rate per hour (0.5).

In wearable biomedical signal detection, the low-power consumption is a critical requirement. However, the process of biomedical signal detection with traditional neural-network processor is uneconomical for large data movements. A typical solution is the near memory computing (NMC) method, locating more data near the computing engine to save energy, where the detecting accuracy and power consumption is difficult to be optimized simultaneously. In addition, a suitable computing engine is needed to match both power and computation budget. In this work, we combine the NMC-based ECG processor equipped with a high-precision time-domain engine to perform the detection of arrhythmia, namely TDPRO. The proposed TDPRO supports high precision multiplication and addition operation with 8-bit input and weight parameters. Also, we propose TD-zero-jumping and idle-shutdown technique to further reduce 63%$\sim$ 91% power consumption of the time-domain engine. The error rate of 8-bit MAC operation in the TDPRO is 1.18%, which is suitable for the ECG detection.

Recent development of the processing-in-memory (PIM) architecture has demonstrated high efficiency by reducing data movements. However, the performance of the conventional PIM architecture is limited by several issues, including frequent bit-line operations, complicated control of data flow, and massive inter-macro data movements. In addition, both analog- and digital-PIM solutions have obstacles to meet requirement of high-precision computation. In this work, we explore the tradeoff between data movement and energy efficiency of PIM architecture. We develop a PIM architecture, namely ReverSearch, to accelerate multiple-and-accumulate operation, equipped with reverse searching engine and look up table operations. Also, the corresponding data mapping and data flow methods are provided to improve the performance of the ReverSearch architecture. Based on our evaluation, ReverSearch improves the energy efficiency by 17.26 × and 3.68 ×, compared to the baseline of LUT-Cache [1] and LAcc [2].

Simultaneous Localization and Mapping (SLAM) is used to solve the problem of autonomous localization and navigation of mobile robots in unknown environments. Loop closure detection is a key part of SLAM, which largely determines accuracy and stability of SLAM. In recent years, some experiments have proved that the loop closure detection system based on neural network is superior to the traditional loop closure detection in both accuracy and real-time performance. In this paper, we propose an adaptive real-time loop closure detection (AR-Loop) method based on monocular vision. A pre-trained convolutional neural network (CNN) is used to extract image features. Then features of different layers are concatenated as image descriptors. In addition, the adaptive candidate matching range algorithm and image-to- sequence calibration algorithm are proposed to improve the performance of the algorithm. Extensive experiments have been conducted on several open datasets to validate the performance of AR-Loop. It has been demonstrated that the recall rate is increased by over 18% compared with other state-of-the-art algorithms when the precision is 100%.

Human activity recognition has been widely used in healthcare applications such as elderly monitoring, exercise supervision, and rehabilitation monitoring. Compared with other approaches, sensor-based wearable human activity recognition is less affected by environmental noise and therefore is promising in providing higher recognition accuracy. However, one of the major issues of existing wearable human activity recognition methods is that although the average recognition accuracy is acceptable, the recognition accuracy for some activities (e.g., ascending stairs and descending stairs) is low, mainly due to relatively less training data and complex behavior pattern for these activities. Another issue is that the recognition accuracy is low when the training data from the test subject are limited, which is a common case in real practice. In addition, the use of neural network leads to large computational complexity and thus high power consumption. To address these issues, we proposed a new human activity recognition method with two-stage end-to-end convolutional neural network and a data augmentation method. Compared with the state-of-the-art methods (including neural network based methods and other methods), the proposed methods achieve significantly improved recognition accuracy and reduced computational complexity.

Monocular visual odometry (VO) is one of the most practical ways in vehicle autonomous positioning, through which a vehicle can automatically locate itself in a completely unknown environment. Although some existing VO algorithms have proved the superiority, they usually need another precise adjustment to operate well when using a different camera or in different environments. The existing VO methods based on deep learning require few manual calibration, but most of them occupy a tremendous amount of computing resources and cannot realize real-time VO. We propose a highly real-time VO system based on the optical flow and DenseNet structure accompanied with the inertial measurement unit (IMU). It cascade the optical flow network and DenseNet structure to calculate the translation and rotation, using the calculated information and IMU for construction and self- correction of the map. We have verified its computational complexity and performance on the KITTI dataset. The experiments have shown that the proposed system only requires less than 50% computation power than the main stream deep learning VO. It can also achieve 30% higher translation accuracy as well.

Zero Velocity Update (ZUPT) has played a key role in Pedestrian Dead Reckoning (PDR) with inertial measurement units (IMU). However, it is both crucial and difficult to determine ZUPT conditions given complex and varying motion types such as walking, fast walking or running, and different walking habits of distinct people, which have direct and significant impact on the tracking accuracy. In this research we proposed a model based on deep neural networks to determine moments when the ZUPT should be conducted. The proposed model ensures nearly identical performance regardless of different motion types. It has been demonstrated by extensive experiments conducted in three different scenarios that our model can work equally well with different pedestrians and walking patterns, enabling the wide use of PDR in real-world applications.

Inertial navigation is a fundamental method for pervasive indoor tacking and navigation. Although PDR based on inertial navigation can achieve robust indoors and outdoors positioning, the positioning accuracy does not meet the accuracy we need, due to the error divergence of the system. We present ZUPT with Kalman filter, a precise, robust technique tracks well even when presented with very noisy sensor data. Key to our ZUPT is zero velocity detection, the step to determine if the person's foot is in stance phase during walking. We used three different methods to detect zero velocity moments and compare their accuracy. Finally, we found that ZUPT using asymptotic zero velocity detection greatly improved the accuracy of inertial navigation. We believe that such a convergent and high precision approach will improve the application of inertial navigation in indoor positioning.

Much attention has been paid to efficiently use channel assignment algorithm to enhance throughput of multi-radio multi-channel (MRMC) wireless Mesh networks (WMNs). However, most prior works on channel assignment assume that all links work on a same rate. In this paper, a multi-rate based channel assignment algorithm called Multi-Rate Based Channel Assignment (MRBCA) is proposed to solve problems caused by multi-rate links. Firstly a proper root node is selected and then a multi-rate topology is established with consideration of connectivity and avoidance of single-direction links. After that, links are ordered and bandwidth loss function is proposed. At last, channels are assigned to links according to the order and minimization of bandwidth loss. Simulation results using ns-3 show that MRBCA outperforms other two classic channel assignment algorithms.