Crowd Counting with respect to Age and Gender by using Faster R-CNN based Detection

Masses gather on a daily basis in events like hajjs, shows, marathons, anniversaries, memorials, festivities, political events, protests and shopping malls. Visual understanding of crowd places is an emerging field in computer vision. Counting the people, based on gender, and estimation of their age in crowd is a tricky job but it has extensive applicability in surveillance, management and planning. Crowd counting is also vital to measure political worth of protests and other unwanted gatherings. In this work, we propose an innovative technique to solve the challenges of crowd counting, gender recolonization, age estimation and localization of people in visual frames. Faster R-CNN is trained to solve these four problems simultaneously. As the task of localization involves the annotations of high-quality frames, we introduce CVML-OCC dataset
¹
that incapacitates the limitations of already available datasets. Alongside, we provide evaluation procedures and comprehensive comparison with existing approaches. Our method expressively performs better than state-of-the-art modules with the training set of CVML-OCC dataset, which is a comprehensive dataset having precise annotations in diverse scenarios.