Search

Intelligent Transportation System (ITS) aims to strengthen the connection between vehicles, roads, and people. As the important road information in ITS, intelligent detection of traffic signs has become an important part in the intelligent vehicle. In this paper, a lightweight vehicle mounted multi-scale traffic sign detector is proposed. First, guided by the attention fusion algorithm, an improved feature pyramid network is proposed, named AFPN. Assign weights according to the importance of information and fuse multi-dimensional attention maps to improve feature extraction and information retention capabilities. Second, a multi-head detection structure is designed to improve the positioning and detection capability of the detector. According to the target scale, the corresponding detection head is constructed to improve the target detection accuracy. The experimental results show that compared with other state-of-the-art methods, the proposed method not only has excellent detection accuracy with 50.3% for small targets and 64.8% for large targets but also can better trade-off detection speed and detection accuracy. Furthermore, the proposed detector is deployed on the Jetson Xavier NX and integrated with the vehicle-mounted camera, inverter, and LCD to realize real-time traffic sign detection on the vehicle terminal, and the speed reaches 25.6 FPS. [ABSTRACT FROM AUTHOR]

Traffic sign detection is a challenging task for the unmanned driving system, especially for the detection of multi-scale targets and the real-time problem of detection. In the traffic sign detection process, the scale of the targets changes greatly, which will have a certain impact on the detection accuracy. Feature pyramid is widely used to solve this problem, but due to the diversity of traffic sign sizes, it cannot accurately extract multi-size feature maps, thus destroying the feature consistency between traffic signs. Moreover, in practical application, it is difficult for common methods to improve the detection accuracy of multi-scale traffic signs while ensuring real-time detection. In this paper, we propose an improved feature pyramid model, named AF-FPN, which utilizes the adaptive attention module (AAM) and feature enhancement module (FEM) to reduce the information loss in the process of feature map generation and enhance the representation ability of the feature pyramid. We replaced the original feature pyramid network in YOLOv5 with AF-FPN, which improves the detection performance for multi-scale targets of the YOLOv5 network under the premise of ensuring real-time detection. Furthermore, a new automatic learning data augmentation method is proposed to enrich the dataset and improve the robustness of the model to make it more suitable for practical scenarios. Extensive experimental results on the Tsinghua-Tencent 100 K (TT100K) dataset demonstrate that compared with several state-of-the-art methods, our method is more universal and superior. [ABSTRACT FROM AUTHOR]

Visual target tracking remains a hard research problem due to many unpredictable challenges (e.g., complete occlusion, out-of-view, and fast motion). This paper proposes a novel kernelized correlation filter architecture based on the fusion of the multiple feature response maps (FRMs), enhanced target re-detection, and improved online model updating. First, the multi-feature fusion strategy reveals that complementary feature information paves the way for effectively improving the target recognition ability. Meanwhile, a novel target re-detection module performs superior in relocating the tracked target and mitigating the model drift problem. Moreover, most existing trackers may indiscriminately update the model when encountering inaccurate predictions, eventually induce undesired model drift and tracking performance decay. Inspiring by the average peak-to-correlation energy, we apply an improved model updating scheme to ascertain the relative fluctuated level of FRMs. This novel model updating method can reject deteriorated samples and improves the discriminative power of the model. As evaluated on the four popular tracking benchmark datasets (including OTB-2013, OTB-2015, UAV123, and LaSOT), benefiting from proposed methods, our tracker obtains better effectiveness and efficiency than that of 28 state-of-the-art competitors (including 17 correlation filter-based competitors and 11 deep learning-based competitors). Meanwhile, our proposed tracker can meet the requirements of limited storage space and real-time performance. [ABSTRACT FROM AUTHOR]

Memristor is a nanoscale circuit element with nonvolatile, binary, multilevel and analog states. Its conductance (resistance) plasticity is similar to biological synapses. Information sparse coding is considered as the key mechanism of biological neural systems to process mass complex perception data, which is applied in the fields of signal processing, computer vision and so on. This paper proposes a soft-threshold adaptive sparse coding algorithm named MMN-SLCA based on the memristor, neural network and sparse coding theory. Specifically, the memristor crossbar array is used to realize the dictionary set. And by leveraging its unique vector–matrix operation advantages and biological synaptic characteristic, two key compositions of the sparse coding, namely, pattern matching and lateral neuronal inhibition are realized conveniently and efficiently. Besides, threshold variability further enhances the adaptive ability of the intelligent sparse coding. Furthermore, a hardware implementation framework of the sparse coding algorithm is designed to provide feasible solutions for hardware acceleration, real-time processing and embedded applications. Finally, the application of MMN-SLCA in image super-resolution reconstruction is discussed. Experimental simulations and result analysis verify the effectiveness of the proposed scheme and show its superior potentials in large-scale low-power intelligent information coding and processing. [ABSTRACT FROM AUTHOR]

A novel systematic design of associative memory networks is addressed in this paper, by incorporating both the biological small-world effect and the recently acclaimed memristor into the conventional Hopfield neural network. More specifically, the original fully connected Hopfield network is diluted by considering the small-world effect, based on a preferential connection removal criteria, i.e., weight salience priority. The generated sparse network exhibits comparable performance in associative memory but with much less connections. Furthermore, a hardware implementation scheme of the small-world Hopfield network is proposed using the experimental threshold adaptive memristor (TEAM) synaptic-based circuits. Finally, performance of the proposed network is validated by illustrative examples of digit recognition. [ABSTRACT FROM AUTHOR]