Abstract
A difficult challenge in computer vision is deciphering human activity from videos. The primary capability of smart video surveillance systems is the automatic identification of human actions in the recorded sequence and the tagging of such actions. The aim of recognition of human behavior is to recognize the activities and purposes of any number of things through a series of investigations into those actions and those of the objects' surroundings. With the emergence of vast quantities of human-centric video data due to technological advancements, the efficient identification of human action from huge video data has become a bottleneck in video processing.
Occlusion and pixels bending moment effects make it difficult for object detection algorithms to work properly, which increases complexity and erroneous margin in addition. In this paper, characteristics are extracted using Apache Spark using in-memory computation and the distributed environment's goal of action by humans recognition. Additionally, the local movement descriptor is utilized to derive features that compute the intensity of the pixel information between two frames, taking both motion and appearance into account. To distinguish human actions, the spark Machine Learning Library random forest is then used. To investigate the model's capacity for activity recognition, experimental tests are carried out, and the outcomes are contrasted with widely used Decision Trees for supervised categorization. In the comparative analysis, it was found that the suggested approach performed better than the other categorization methods.
