IMPROVING MULTI-CAMERA ACTIVITY RECOGNITION BY EMPLOYING NEURAL NETWORK BASED READJUSTMENT

摘要: In this paper, we propose a method to enhance activity recognition in complex environments, where problems like occlusions, outliers and illumination changes occur. order address the induced by dependency on camera's viewpoint, multiple cameras are used an endeavor exploit redundancies. We initially examine effectiveness of various information stream fusion approaches based hidden Markov models, including Student's t-endowed models for tolerance outliers. Following, introduce neural network-based readjustment mechanism that fits these schemes aims at dynamically correcting erroneous classification results image sequences, thus improving overall rates. The proposed evaluated under real life scenarios, acquired compared discussed.