eduzhai > Applied Sciences > Engineering >

Optimisation of a Siamese Neural Network for Real-Time Energy Efficient Object Tracking

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: In this paper the research on optimisation of visual object tracking using aSiamese neural network for embedded vision systems is presented. It was assumedthat the solution shall operate in real-time, preferably for a high resolutionvideo stream, with the lowest possible energy consumption. To meet theserequirements, techniques such as the reduction of computational precision andpruning were considered. Brevitas, a tool dedicated for optimisation andquantisation of neural networks for FPGA implementation, was used. A number oftraining scenarios were tested with varying levels of optimisations - frominteger uniform quantisation with 16 bits to ternary and binary networks. Next,the influence of these optimisations on the tracking performance was evaluated.It was possible to reduce the size of the convolutional filters up to 10 timesin relation to the original network. The obtained results indicate that usingquantisation can significantly reduce the memory and computational complexityof the proposed network while still enabling precise tracking, thus allow touse it in embedded vision systems. Moreover, quantisation of weights positivelyaffects the network training by decreasing overfitting.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×