Probabilistic models for concurrent chatting activity recognition

Abstract

Recognition of chatting activities in social interactions is useful for constructing human social networks. However, the existence of multiple people involved in multiple dialogues presents special challenges. To model the conversational dynamics of concurrent chatting behaviors, this paper advocates Factorial Conditional Random Fields (FCRFs) as a model to accommodate co-temporal relationships among multiple activity states. In addition, to avoid the use of inefficient Loopy Belief Propagation (LBP) algorithm, we propose using Iterative Classification Algorithm (ICA) as the inference method for FCRFs. We designed experiments to compare our FCRFs model with two dynamic probabilistic models, Parallel Condition Random Fields (PCRFs) and Hidden Markov Models (HMMs), in learning and decoding based on auditory data. The experimental results show that FCRFs outperform PCRFs and HMM-like models. We also discover that FCRFs using the ICA inference approach not only improves the recognition accuracy but also takes significantly less time than the LBP inference method.

GARSAaaS: group activity recognition and situation analysis as a service

“…The barometer sensor data indicates atmospheric pressure and we can map atmospheric pressure to an elevation value. To do this, barometer data from four smartphones were collected via 4 participants in one hour of normal walking in a place 20 with hills (up and downs) which shows barometer changes more clearly. Every 10 s, the elevation of participants were also captured by use of Google's elevation API 21 and the MapQuest API 22 .…”

Interaction-Feature Enhanced Multiuser Model Learning for a Home Environment Using Ambient Sensors

“…For example, a user can use a wearable camera to detect the posture of the other person interacting with him/her, or can use a wearable microphone to detect the content of the conversation to the other. 39 In summation, for building activity models, we define the following three feature sets: "primitive," "obj_loc," and "obj_loc_motion-off." The elements in these feature sets are {event, interaction}, {object, location, interaction}, and {object, location, motion-off-location, interaction}, respectively.…”

Recognizing Group Activities Using Wearable Sensors

“…Further work in group activity recognition was conducted using camera-based activity recognition, such as for automatically recognizing group activities in a prison yard [5]. Another approach uses audio classification to recognize group activities, such as concurrent chatting activities [11], or for classifying roles of individuals in conversations and meetings [6]. These methods have proven effective, but rely heavily on infrastructure for recognition (cameras, networks, etc.).…”