Recent Progress in Improving the Performance of Infrared Photodetectors via Optical Field Manipulations

Benefiting from the inherent capacity for detecting longer wavelengths inaccessible to human eyes, infrared photodetectors have found numerous applications in both military and daily life, such as individual combat weapons, automatic driving sensors and night-vision devices. However, the imperfect material growth and incomplete device manufacturing impose an inevitable restriction on the further improvement of infrared photodetectors. The advent of artificial microstructures, especially metasurfaces, featuring with strong light field enhancement and multifunctional properties in manipulating the light–matter interactions on subwavelength scale, have promised great potential in overcoming the bottlenecks faced by conventional infrared detectors. Additionally, metasurfaces exhibit versatile and flexible integration with existing detection semiconductors. In this paper, we start with a review of conventionally bulky and recently emerging two-dimensional material-based infrared photodetectors, i.e., InGaAs, HgCdTe, graphene, transition metal dichalcogenides and black phosphorus devices. As to the challenges the detectors are facing, we further discuss the recent progress on the metasurfaces integrated on the photodetectors and demonstrate their role in improving device performance. All information provided in this paper aims to open a new way to boost high-performance infrared photodetectors.

Mental workLoad Accumulation Effect of Mobile Phone Distraction in L2 Autopilot Mode

As self-driving vehicles become more common, there is a need for precise measurement and definition of when and in what ways a driver can use a mobile phone in autonomous driving mode, for how long it can be used, the complexity of the call content, and the accumulated psychological load. This study uses a 2 (driving mode) * 2 (call content complexity) * 6 (driving phase) three-factor mixed experimental design to investigate the effect of these factors on the driver's psychological load by measuring the driver's performance on peripheral visual detection tasks, pupil diameter, and EEG components in various brain regions in the alpha band. The results showed that drivers' mental load levels converge between manual and automatic driving modes as the duration of driving increases, regardless of the level of complexity of the mobile phone conversation. This suggests that mobile phone conversations can also disrupt the driver's cognitive resource balance in automatic driving mode, as it increases mental load while also impairing the normal functioning of brain functions such as cognitive control, problem solving, and judgment, thereby compromising driving safety.

Research on quantum cognition in autonomous driving

Autonomous vehicles for the intention of human behavior of the estimated traffic participants and their interaction is the main problem in automatic driving system. Classical cognitive theory assumes that the behavior of human traffic participants is completely reasonable when studying estimation of intention and interaction. However, according to the quantum cognition and decision theory as well as practical traffic cases, human behavior including traffic behavior is often unreasonable, which violates classical cognition and decision theory. Based on the quantum cognitive theory, this paper studies the cognitive problem of pedestrian crossing. Through the case analysis, it is proved that the Quantum-like Bayesian (QLB) model can consider the reasonability of pedestrians when crossing the street compared with the classical probability model, being more consistent with the actual situation. The experiment of trajectory prediction proves that the QLB model can cover the edge events in interactive scenes compared with the data-driven Social-LSTM model, being more consistent with the real trajectory. This paper provides a new reference for the research on the cognitive problem of intention on bounded rational behavior of human traffic participants in autonomous driving.

Map matching for circular road via contextual voting

Due to the limitation of positioning devices, there is a certain error between GPS positioning data and the real location on the map, and the positioning data needs to be processed to have better usability. For example, accurate location is needed for traffic flow control, automatic driving navigation, logistics tracking, etc. There are few studies specifically for circular road sections. In addition, many existing map matching methods based on Hidden Markov model (HMM) also have the problem that GPS points are easily to be matched to tangent or non-adjacent road sections at circular road sections. Therefore, the contextual voting map matching method for circular road sections (STDV-matching) is proposed. The method proposes multiple subsequent point direction analysis methods based on STD-matching to determine entry into the circular section, and adds candidate section frequency voting analysis to reduce matching errors. The effectiveness of the proposed method is verified at the circular section by comparing it with three existing HMM methods through experiments using two real map and trajectory datasets.

Semantic segmentation of large-scale point clouds based on dilated nearest neighbors graph

Three-dimensional (3D) semantic segmentation of point clouds is important in many scenarios, such as automatic driving, robotic navigation, while edge computing is indispensable in the devices. Deep learning methods based on point sampling prove to be computation and memory efficient to tackle large-scale point clouds (e.g. millions of points). However, some local features may be abandoned while sampling. In this paper, We present one end-to-end 3D semantic segmentation framework based on dilated nearest neighbor encoding. Instead of down-sampling point cloud directly, we propose a dilated nearest neighbor encoding module to broaden the network’s receptive field to learn more 3D geometric information. Without increase of network parameters, our method is computation and memory efficient for large-scale point clouds. We have evaluated the dilated nearest neighbor encoding in two different networks. The first is the random sampling with local feature aggregation. The second is the Point Transformer. We have evaluated the quality of the semantic segmentation on the benchmark 3D dataset S3DIS, and demonstrate that the proposed dilated nearest neighbor encoding exhibited stable advantages over baseline and competing methods.

Improved YOLO v5 with balanced feature pyramid and attention module for traffic sign detection

With the development of automatic driving technology, traffic sign detection has become a very important task. However, it is a challenging task because of the complex traffic sign scene and the small size of the target. In recent years, a number of convolutional neural network (CNN) based object detection methods have brought great progress to traffic sign detection. Considering the still high false detection rate, as well as the high time overhead and computational overhead, the effect is not satisfactory. Therefore, we employ lightweight network model YOLO v5 (You Only Look Once) as our work foundation. In this paper, we propose an improved YOLO v5 method by using balances feature pyramid structure and global context block to enhance the ability of feature fusion and feature extraction. To verify our proposed method, we have conducted a lot of comparative experiments on the challenging dataset Tsinghua-Tencent-100K (TT100K). The experimental results demonstrate that the [email protected] and [email protected]:0.95 are improved by 1.9% and 2.1%, respectively.

Object Detection in Autonomous Driving Scenarios Based on an Improved Faster-RCNN

Object detection is one of the key algorithms in automatic driving systems. Aiming at addressing the problem of false detection and the missed detection of both small and occluded objects in automatic driving scenarios, an improved Faster-RCNN object detection algorithm is proposed. First, deformable convolution and a spatial attention mechanism are used to improve the ResNet-50 backbone network to enhance the feature extraction of small objects; then, an improved feature pyramid structure is introduced to reduce the loss of features in the fusion process. Three cascade detectors are introduced to solve the problem of IOU (Intersection-Over-Union) threshold mismatch, and side-aware boundary localization is applied for frame regression. Finally, Soft-NMS (Soft Non-maximum Suppression) is used to remove bounding boxes to obtain the best results. The experimental results show that the improved Faster-RCNN can better detect small objects and occluded objects, and its accuracy is 7.7% and 4.1% respectively higher than that of the baseline in the eight categories selected from the COCO2017 and BDD100k data sets.