Positive developmental video classification for children

Abstract

This paper introduces the concept of positive developmental video classification. The work focuses on developing features and classification systems that can be used to classify content based on the impact on the cognitive, social and academic development of children according to an expertly assigned predefined positive or negative cognitive impact category. We solve the problem by developing novel features that gauge the amount of social interaction, attention disrupting fast-paced content, incorporate music information retrieval features and combine these features with other video content analysis features. This information is then used to determine what content has a positive impact on a child's development. It was found that the low-level features can be used for classification and do have correlation with expertly assigned predefined impact categories. To ensure the validation results are not based on similarities between content, a new model validation technique is developed to ensure that the videos are classified with respect to their impact on development. In addition, we developed a data set of videos that has been classified as having a positive or negative impact on children, based on expert experimental results in the psychological literature. This data set can be used as a benchmark for future research. Validation results found the system had almost 30% better accuracy than state-of-the-art video genre classification systems and over 65% better performance than the arousal time curve used in affective video content modelling.