The aim of this paper is to find the optimal design for tiny YOLO2 in an embedded system environment. Our focus is to rebuild the given YOLO2 code and to find the optimal design parameters in order to maximize the spe...
详细信息
ISBN:
(纸本)9789813292444;9789813292437
The aim of this paper is to find the optimal design for tiny YOLO2 in an embedded system environment. Our focus is to rebuild the given YOLO2 code and to find the optimal design parameters in order to maximize the speed using the light-weight GPGPU in a target embedded environment. To maximize the YOLO2 performance we used OpenCL framework while utilizing the embedded GPGPU and tried various aspects of OpenCL design parameters such work item, work group, and resulting in reducing the global memory access overhead and maximizing computing load balancing between computing units under constraints including local memory resources and computing resources. Our experimental results show that the overall performance enhancement is 18.2 times compared to the naive implementation.
The applications of UAVs (unmanned aerial vehicles) have been increasing and becoming part of many daily tasks in numerous organizations. As matter of fact, the use of a UAV does not mean the decreasing of operational...
详细信息
The applications of UAVs (unmanned aerial vehicles) have been increasing and becoming part of many daily tasks in numerous organizations. As matter of fact, the use of a UAV does not mean the decreasing of operational complexities and, consequently, the costs of performing its tasks. Sometimes, this high cost is related to the dependence of well-trained operators and huge remote control facilities to operate a sophisticated UAV. This work proposes an UAV that can perform its tasks as much independent of human interaction as possible, and with a minimum connection to its mission control facilities. This independence will be achieved by embedding the mission control into the UAV. As the mission control is embedded, the UAV will have less connection issues with its control center and will be less dependable of human interaction. To prove this concept, the kinematics and dynamics of a light air vehicle (blimp) were developed;a prototype of an embedded parallel-distributed computer was constructed;and new procedures to resolve navigations and collision evasions issues were proposed. The new evasion procedures were implemented into a simulator and a new parallel/distributed program for optimal path discover was developed to be used in the cluster prototype. All tests of the evasion procedures simulator were satisfactory and the speed up tests using the embedded cluster showed the best performance of the proposed framework.
the applications of UAVs (Unmanned Aerial Vehicles) are growing up and becoming part of many daily tasks in many organizations. However, as matter of fact, the use of a UAV does not mean the decreasing of operational ...
详细信息
ISBN:
(纸本)9781467371292
the applications of UAVs (Unmanned Aerial Vehicles) are growing up and becoming part of many daily tasks in many organizations. However, as matter of fact, the use of a UAV does not mean the decreasing of operational complexities and, consequently, the costs of perform its tasks. Some times this high cost is connected to the dependence of well-trained operators and huge remote control facilities to operate a sophisticated UAV's. This paper proposes a robotic flying crane using UAV that can perform its tasks as much independent of human interaction as possible, and with a minimum connection to any mission control facilities. This independence will be achieved by embedding the path mission control into the UAV. As path control is embedded, the UAV will have less connections issues with its control center and will be less dependable of human interaction. To prove this concept all kinematics and dynamics of a light than air vehicle (blimp) is develop;a prototype of an embedded parallel-distributed compute was constructed;and new procedures to resolve navigations and collision evasions issues are proposed. The new evasion procedures were implemented into a simulator and a new parallel / distributed program for optimal path discover was developed to be used into the cluster prototype. Tests of the evasion procedures simulator were satisfactory and speed up of embedded proposed cluster showed better performance of proposed framework.
暂无评论