A MACHINE LEARNING BASED ARCHITECTURE IMPLEMENTATION FOR OBJECT DETECTION

  • S.Mohanraj et al.

Abstract

The ConvNet (CNN) is a neural network technique that has been extensively employed in
detect and recognize object, video analytics and image classification. ConvNet based algorithms
aren’t easy to be deployed in the embedded devices as the ConvNet is memory expensive and
compute-intensive. There are many studies that have reviewed the implementation of ConvNet based
classification models on hardware, a vast area of implementation ConvNet on FPGA is yet to be
explored. This work proposed the half-precision floating point ConvNet implementation to detect
object with Tiny YOLO v2 on Intel Altera Cyclone-IV FPGA using Verilog HDL. In view of the
resource constrains of an FPGA in relationswith memory bandwidth, computational resources along
with on-chip memory, the pre-processing of data is planned so that the concatenationof batch
normalization addicted to convolution layer.

Published
2020-02-16
Section
Articles