SEMANTIC SEGMENTATION USING OPTICAL SENSORS IN THE TASK OF AUTONOMOUS DRIVING

This paper is devoted to the problem of image semantic segmentation for machine vision system of off-road autonomous robotic vehicle. Most modern convolutional neural networks require large computing resources that go beyond the capabilities of many robotic platforms. Therefore, the main drawback of such models is extremely high complexity of the convolutional neural network used, whereas tasks in real applications must be performed on devices with limited resources in real-time. This paper focuses on the practical application of modern lightweight architectures as applied to the task of semantic segmentation on mobile robotic systems. The article discusses backbones based on ResNet18, ResNet34, MobileNetV2, ShuffleNetV2, EfficientNet-B0 and decoders based on U-Net, DeepLabV3 and DeepLabV3+ as well as additional components that can increase the accuracy of segmentation and reduce the inference time. In this paper we propose a model using ResNet34 enconding and DeepLabV3+ decoding with Squeeze & Excitation blocks that was optimal in terms of inference time and accuracy. We also demonstrate our off-road dataset and simulated dataset for semantic segmentation. Furthermore, we increased mIoU metric by 2.6 % on our off-road dataset using pretrained weights on simulated dataset, compared with mIoU metric using pretrained weights on the Cityscapes. Moreover, we achieved 76.1 % mIoU on the Cityscapes validation set and 85.4 % mIoU on our off-road validation set at 37 FPS (Frames per Second) for an input image of 1024×1024 size on one NVIDIA GeForce RTX 2080 card using NVIDIA TensorRT inference framework.

Tutorial: How to Generate Missing Data For Simulation Studies

Missing data are common in psychological and educational research. With the improvement in computing technology in recent decades, more researchers begin developing missing data techniques. In their research, they often conduct Monte Carlo simulation studies to compare the performances of different missing data techniques. During such simulation studies, researchers must generate missing data in the simulated dataset by deciding which data values to delete. However, in the current literature, there are few guidelines on how to generate missing data for simulation studies. Our paper is one of the first papers that examines ways of generating missing data for simulation studies. We emphasize the importance of specifying missing data rules which are statistical models for generating missing data. We begin the paper by reviewing the types of missing data mechanisms and missing data patterns. We then explain how to specify missing data rules to generate missing data with different mechanisms and patterns. We end the paper by presenting recommendations for generating missing data for simulation studies.

UAV Detection with Transfer Learning from Simulated Data of Laser Active Imaging

With the development of our society, unmanned aerial vehicles (UAVs) appear more frequently in people’s daily lives, which could become a threat to public security and privacy, especially at night. At the same time, laser active imaging is an important detection method for night vision. In this paper, we implement a UAV detection model for our laser active imaging system based on deep learning and a simulated dataset that we constructed. Firstly, the model is pre-trained on the largest available dataset. Then, it is transferred to a simulated dataset to learn about the UAV features. Finally, the trained model is tested on real laser active imaging data. The experimental results show that the performance of the proposed method is greatly improved compared to the model not trained on the simulated dataset, which verifies the transferability of features learned from the simulated data, the effectiveness of the proposed simulation method, and the feasibility of our solution for UAV detection in the laser active imaging domain. Furthermore, a comparative experiment with the previous method is carried out. The results show that our model can achieve high-precision, real-time detection at 104.1 frames per second (FPS).

An Algorithm for Choosing the Optimal Number of Muscle Synergies during Walking

In motor control studies, the 90% thresholding of variance accounted for (VAF) is the classical way of selecting the number of muscle synergies expressed during a motor task. However, the adoption of an arbitrary cut-off has evident drawbacks. The aim of this work is to describe and validate an algorithm for choosing the optimal number of muscle synergies (ChoOSyn), which can overcome the limitations of VAF-based methods. The proposed algorithm is built considering the following principles: (1) muscle synergies should be highly consistent during the various motor task epochs (i.e., remaining stable in time), (2) muscle synergies should constitute a base with low intra-level similarity (i.e., to obtain information-rich synergies, avoiding redundancy). The algorithm performances were evaluated against traditional approaches (threshold-VAF at 90% and 95%, elbow-VAF and plateau-VAF), using both a simulated dataset and a real dataset of 20 subjects. The performance evaluation was carried out by analyzing muscle synergies extracted from surface electromyographic (sEMG) signals collected during walking tasks lasting 5 min. On the simulated dataset, ChoOSyn showed comparable performances compared to VAF-based methods, while, in the real dataset, it clearly outperformed the other methods, in terms of the fraction of correct classifications, mean error (ME), and root mean square error (RMSE). The proposed approach may be beneficial to standardize the selection of the number of muscle synergies between different research laboratories, independent of arbitrary thresholds.

Long-Run Effects in Dynamic Systems: New Tools for Cross-Lagged Panel Models

Cross-lagged panel models (CLPMs) are common, but their applications often focus on “short-run” effects among temporally proximal observations. This addresses questions about how dynamic systems may immediately respond to interventions, but fails to show how systems evolve over longer timeframes. We explore three types of “long-run” effects in dynamic systems that extend recent work on “impulse responses,” which reflect potential long-run effects of one-time interventions. Going beyond these, we first treat evaluations of system (in)stability by testing for “permanent effects,” which are important because in unstable systems even a one-time intervention may have enduring effects. Second, we explore classic econometric long-run effects that show how dynamic systems may respond to interventions that are sustained over time. Third, we treat “accumulated responses” to model how systems may respond to repeated interventions over time. We illustrate tests of each long-run effect in a simulated dataset and we provide all materials online including user-friendly R code that automates estimating, testing, reporting, and plotting all effects (see https://doi.org/10.26188/13506861 ). We conclude by emphasizing the value of aligning specific longitudinal hypotheses with quantitative methods.