Search

Simultaneous Localization and Mapping (SLAM) systems face high computational demands, hindering their real-time implementation on low-end computers. An approach to addressing this challenge involves offline processing, i.e., a map of the environment map is created offline on a powerful computer and then passed to a low-end computer, which uses it for navigation, which involves fewer resources. However, even creating the map on a powerful computer is slow since SLAM is designed as a sequential process. This work proposes a parallel mapping method pSLAM for speeding up the offline creation of maps. In pSLAM, a video sequence is partitioned into multiple subsequences, with each processed independently, creating individual submaps. These submaps are subsequently merged to create a unified global map of the environment. Our experiments across a diverse range of scenarios demonstrate an increase in the processing speed of up to 6 times compared to that of the sequential approach while maintaining the same level of robustness. Furthermore, we conducted comparative analyses against state-of-the-art SLAM methods, namely UcoSLAM, OpenVSLAM, and ORB-SLAM3, with our method outperforming these across all of the scenarios evaluated. [ABSTRACT FROM AUTHOR]

Convolution operations lead to a significant increase in the network size, which makes CNN models difficult to deploy to the embedded hardware platform, and different granularity data is not coordinated with the underlying hardware structure, which leads to low computing efficiency. Based on the reconfigurable array with the computing units supporting multiple bit widths, through software hardware cooperation and reconfigurable computing methods, this paper defined the quantization threshold using KL divergence and random integer method, proposed a strategy for finding the best basis point, designed an instruction set and a parallel mapping scheme supporting multiple bit widths to realize three distinct bit widths in data quantization. The results show the quantization scheme with 8 bit weight and feature map can compress model parameter quantity to about 50% with 2% accuracy loss. The acceleration ratios of quantifying the test images to three different bit widths reach 1.012, 1.273, and 1.556, respectively, which can shorten the execution time by up to 35.7% and reduce memory access times by 56.2%, while only bringing less than 1% relative error. This indicates that this method can achieve efficient neural network computation under three quantization bit widths, thereby implementing hardware acceleration and model compression. [ABSTRACT FROM AUTHOR]

A parallel real-time visualization software tool HiVecMap is proposed here for large-scale geographic vector data mapping. HiVecMap calculates the mapping value of each display cell instead of the value of each vector feature. The mapping value of each display cell is determined by the number of geographic vector features in the corresponding spatial range. Thereby, the computational complexity is depends on the number of the cells rather than the number of features of the whole geographic dataset. When the number of the cells is much smaller than the number of geographic features, the visualization efficiency can be greatly improved. HiVecMap is implemented with the MPI+OpenMP parallel strategy, and has been extensively tested. Results show that large-scale geographic vector data can be online mapped in real-time, which has a performance improvement of more than ten times compared with the existing best methods.

Competing Interests: The authors have no conflicts of interest to disclose.

A parallel mapping of self-organizing map (SOM) algorithm is presented for a partial tree shape neurocomputer (PARNEU). PARNEU is a general purpose parallel neurocomputer that is designed for soft computing applications. Practical scalability and a reconfigurable partial tree network are the main architectural features. The presented neuron parallel mapping of SOM with on-line learning illustrates a parallel winner neuron search and a coordinate transfer that are performed in the partial tree network. Phase times are measured to analyse speedup and scalability of the mapping. The performance of the learning phase in SOM with a four processor PARNEU configuration is about 26 MCUPS and the recall phase performs 30 MCPS. Compared to other mappings done for general purpose neurocomputers, PARNEU's performance is very good. [ABSTRACT FROM AUTHOR]