Air pollution is a major threat to mankind nowadays. Major constituents in the polluted air, i.e., PM2.5(Particulate Matter 2.5) and PM10(Particulate Matter 10) can be estimated using sensors. However, in this work, an algorithm is proposed for estimating the Air Quality Index, PM2.5, and PM10 from the outdoor images captured by a webcam. This problem is considered as a regression problem. Features from the images are computed by using a state-of-the-art self-supervised learning model DinoV2. The extracted features are given to a regressor. Various single output regressor and multiple output regressors are explored. The proposed algorithm is tested on the images collected from the cities in India and Nepal (12,240 images). Size of the feature vector is reduced by using genetic algorithm. For a tenfold cross-validation, the value of R2(Coefficient of determination) for AQI prediction is well above 0.98 for K-neighbourhood regressor.