Near Real-Time Object Recognition for Pepper based on Deep Neural Networks Running on a Backpack

Adapted Pepper

“…However, a time of response of one or two seconds in the interaction with a human is suggested to be more humanlike [8] [9]. Reducing the performances of the used algorithms or improving the hardware to embed the computation [10] can reduce the latency with the benefice of a smoother interaction.…”

“…The integration of a GPU within Pepper has already been studied by the university of Chile [10] and by the university of Salerno [12]. These integrations had the same purpose as Adapted Pepper, to enhance the perception capabilities of the robot, specifically in the fields of:…”

“…• Object recognition • Face detection • Age and gender recognition • Emotion recognition In [10], the GPU was integrated as a Backpack and directly connected to Pepper through an Ethernet cable. The GPU and the camera are further integrated into the Adapted Pepper to answer to design constraints.…”

“…The final module (Fig. 5) to integrate is composed of: To measure the maximum temperature of the camera, the 3D and 2D images are streamed using the official library provided by the constructor 10 during an hour to increase the temperature of the camera. In order to take into account the heat of the top camera of Pepper, the measurements in the configuration 2b and 2c are done with the robot switched on.…”