With the advent of Artificial Intelligence (AI) and automation, Neural Networks are being explored as stateof-the-art algorithms for various tasks that require cognitive abilities that machines do not possess. Few of the most well-known problems in computer vision and AI are those of image classification, object detection, semantic and instance segmentation in images. While humans learn the ability to recognise objects and semantics from experience and with practice, machines interpret images merely as a collection of numbers or pixel values that are meaningless to them. Convolutional Neural Networks (ConvNets) are state-of the- art algorithms used to accomplish various computer vision tasks that require perception. ConvNets attempt to simulate human cognitive abilities of identifying objects by processing information in neurons like visual signals in living beings as they travel through synapses. 
 
The main objective of this project was to design and implement an efficient and ubiquitous FPGA-based ConvNet accelerator that can be applied to any of the modern embedded machine vision applications. Furthermore, performance gained from utilising the designed accelerator would be demonstrated by applying our accelerator to object classification in images. Input data and model quantization schemes were also determined to find a right balance among the accuracy, resource utilization and performance of the designed accelerator.