Multi-Feature Descriptors for Human Interaction Recognition in Outdoor Environments

In recent years, Artificial intelligence has gained significant attention in automatically recognizing human interaction. Thus, leading researchers to develop efficient Human Interaction Recognition (HIR) systems. Efficient feature extraction is the key to developing a system that achieves higher accuracies. Feature extraction emphasizes identifying robust features that can effectively distinguish various human behaviors. This paper introduces a novel approach enabling machines to understand human behaviors in an outdoor environment to recognize human interactions efficiently. This research aims to extract key body points from human silhouettes for robust spatio-temporal features that exhibit unique characteristics for each interaction. Our proposed human interaction recognition (HIR) system recognizes six complex human interactions taken from UT - Interaction dataset i.e. handshake, punch, kick, point, hug, and push. We have designed hybrid feature algorithms and a multiclass support vector machine (SVM) for interaction recognition. To evaluate our system's performance, we have computed Mean recognition accuracy and compared it with other state-of-the-art classifiers. The experimental results showcase the reliability of our proposed approach in handling complex real-world scenarios. Thus, making it applicable to vast domains such as smart homes, violence detection, security systems, and surveillance.