Depth Sensors-Based Action Recognition using a Modified K-Ary Entropy Classifier

Surveillance system is acquiring an ample interest in the field of computer vision. Existing surveillance usually relies on optical or wearable sensors for indoor and outdoor activities. These give reasonable performance a simulation environment. However, when used under realistic settings, they could cause large number false alarms. Moreover, real-world scenario, positioning depth camera at too great distance from subject compromise image quality result loss information. Furthermore, information RGB images may be lost converting 3D to 2D image. Therefore, extensive research moving fused sensors, which has greatly improved action recognition performance. By taking into account concept this paper proposed novel idea modified K-Ary entropy classifier algorithm map arbitrary size vectors fixed-size subtree pattern graph classification solve complex feature selection problems using RGB-D data. The main aim increase space between intra-substructure nodes tree through accumulation. Hence, likelihood classifying minority class as belonging majority been reduced. working model described follows: First, three benchmark datasets have taken input model. Then, 2.5D cloud point modeling ridge extraction, full-body features, point-based features retrieved. Finally, efficacy system, accumulation optimized by probability-based incremental learning (PBIL) used. In both qualitative quantitative experimental results, testing results shown 95.05%, 95.56%, 95.08% over SYSU-ACTION, PRECIS HAR, Northwestern-UCLA (N-UCLA) datasets. apply various emerging applications like human target tracking, security-critical event detection, perimeter security, internet public safety etc.