COVID-19 is a global health emergency that has fundamentally altered human life. Public perception about COVID-19 greatly informs public policymaking and charts the course of present and future mitigation strategies. Existing approaches to gain insights into the evolving nature of public opinion has led to the application of natural language processing on public interaction data acquired from online surveys and social media. In this work, we apply supervised and unsupervised machine learning approaches on global Twitter data to learn the opinions about adoption of mitigation strategies such as social distancing, masks, and vaccination, as well as the effect of socioeconomic, demographic, political, and epidemiological features on perceptions. Our study reveals the uniform polarity in public sentiment on the basis of spatial proximity or COVID-19 infection rates. We show the reservation about the adoption of social distancing and vaccination across the world and also quantify the influence of airport traffic, homelessness, followed by old age and race on sentiment of netizens within the US.