Sensor Free Wireless Activity Recognition for Smart Retail

Activity recognition is an important technology that can be applied to many real-life and human-centric problems. Most techniques for activity recogntion propose sensor-based solutions in which the user carries the sensor and participates actively in recognition. In this paper, we propose a realtime and sensor-free algorithm using low-power and low-cost sensors deployed indoors to recognize the consumer's activity in smart retail scenarios. The proposed sensor-free recognition algorithm is based on the real-time analysis of the wireless signal perturbation induced by the consumer movement in the area covered by sensor nodes. We apply Bayesian technique exploiting the signal fluctuation and variation over time. In order to evaluate the performance of the sensor-free algorithm, experimental measurements are conducted inside an indoor spot to identify consumer activity shopping, including walking in a specific section, dwelling for a while, gazing at the specific item and reaching it out. In addition, the proposed sensor-free algorithm is compared with the sensor-based as the consumer carries wearbale sensor. Experimental results confirm that the estimation accuracy of the proposed approach is comparable with the wearable one, particularly for the walking action.