Audience analysis solution using soft computing methodologies

In this paper, we propose a pedestrian analysis solution helpful for adaptive content delivery and interest measurement for outdoor advertisement displays. The proposed system has built-in camera on the top panel of such displays which capture the real time viewers' frames. The captured frames have been analyzed for detection of faces using Viola-Jones algorithm. The detected faces have been processed with various image processing operations and age, gender, and race (ethnicity) parameters have been estimated using machine learning approaches. The current set of experiments has been generated using Classification and Regression Trees (CART) with Adaboost, Support vector machines (SVMs), Scalable boosting, and neural networks. Dimensionality reduction techniques and other optimizations have been adapted to make the system run efficiently on embedded devices. Computational experiments generated over thousands of real images and live video streams are encouraging and we plan to deploy the solution on large scale advertisement panels.