GWVT: A GPU maritime vessel tracker based on the wisard weightless neural network
Rodrigo da Silva Moreira, Nelson F. F. Ebecken · 2017 Computing Conference · 2017
Maritime surveillance systems increase the security of ports and ships. The video tracking is an important and challenging component of a surveillance system. Difficulties can arise due to weather conditions, target trajectory and appearance, occlusions, lighting conditions and noise. The tracker locates the vessel frame by frame in real time. This paper proposes the GPU WiSARD Vessel Tracker GWVT, a maritime vessel tracker that uses the WiSARD weightless neural network and implemented on a GPU. The GPU parallel processing feature allows the tracking algorithm to be executed very fast. CUDA easy the implementation of the parallel WiSARD tracker because the network discriminators fit well in the thread and block layout. The implementation of a WiSARD based vessel tracker on a GPU is innovative on literature. The GWVT realizes the vessel tracking at only 1 millisecond allowing other tracking techniques to be executed in parallel to rise the performance. Discounting the kernel function call time from GWVT average tracking time, GWVT becomes 13,95 faster than the CPU tracker version.