Automata design for honeybee search algorithm and its applications to 3D scene reconstruction and video tracking

Design of a Low-Power Embedded System Based on a SoC-FPGA and the Honeybee Search Algorithm for Real-Time Video Tracking

“…The honeybee search algorithm (HSA) is similar to other honeybee-inspired meta-heuristics that mimic honeybees’ search for food. The metaphor of flying individuals searching for food is useful for the general image processing problem because the search space is large in both cases and hints of the optimal solution are widely dispersed [ 38 ]. In that sense, HSA is heavily influenced by the dynamic flies heuristic of Boumaza and Louchet [ 53 ], but with a greater emphasis on coordinated work.…”

“…This work applies the findings of that work to propose a workflow for designing another HSA implementation that uses an SoC-PFGA platform in an attempt to overcome the limitations observed with the restrictive CPU-GPU platform. Another recent publication [ 38 ] proposes an alternative representation of HSA using automata and formal languages theory. That representation is the basis for the proposed workflow ( Section 2.5 ), facilitating the proper design process of the real-time video tracking embedded system ( Section 3.1 ).…”

“…These values of the two-dimensional arrays are commonly grayscale pixels stored as integers ranging from 0 to 255 (8-bits). In this work, ZNCC is used as the HSA fitness function because ZNCC remains a reliable, deterministic, non-trained legacy tracker proposal that has been successfully interpreted as an objective function as it has a well-defined domain that can be translated into a search space [ 38 ].…”

“…Moreover, the performance in terms of accuracy is evaluated using the ALOV benchmark in order to confirm that there are no negative effects compared to the employed base similarity measure (ZNCC). This work contributes with (1) the proposed workflow for the design of an embedded system based on an automaton that describes the behavior of a honeybee searching for food [ 38 ] and an SoC-FPGA platform; (2) the design, implementation, and evaluation of a low-power embedded system that performs real-time video tracking using the combination of the HSA meta-heuristic and an SoC-FPGA platform, which was designed from scratch using the co-design methodology [ 39 ]; (3) the comparison of our proposal with a CPU-GPU HSA-based video tracking system [ 34 ] in terms of speed, energy consumption, accuracy, as well as portability; and finally, (4) the contrast with other state-of-the-art video trackers in terms of speed and energy consumption to highlight the advantages of our proposal in these terms.…”

Object Detection and Depth Estimation Approach Based on Deep Convolutional Neural Networks

“…To solve the two problems, several depth estimation approaches were proposed. The conventional stereo vision approaches [ 19 , 20 , 21 , 22 ] included CNN-based methods and video-based processing. Deep learning techniques markedly improved depth estimation performance on the KITTI dataset [ 23 ].…”