RECOGNITION OF HUMAN ACTIVITY BY VIDEO DATA

The paper considers the problem solution of classifying type physical activity a person according to visual data. authors propose using deep neural networks determine activity. recognizing human from video data or single image systems are currently actively used in various areas As example we can take system for monitoring effectiveness enterprise employees. So solving actions is an actual task. developed algorithm determining by based on DenseNet121 and MobileNetV2 models. Then network model was built hyperparameters were selected, because pre-trained did not provide required accuracy detecting software implementation made IDLE environment Python programming language. Experimental studies performed specialized UCF50 dataset containing 50 different types confirm proposed approach solve problem. Additionally, representativeness test set increased with help sequences obtained YouTube.
 Purpose – development person’s data.
 Methodology: work methods computer vision, learning object-oriented used.
 Results: tracking technologies has been developed.
 Practical implications: results be systems, example, criminal activity, medical diagnostics, office employees, etc.