Hand Sign Recognition at Hannover Messe, 30.5.-2.6.2022
Hand Sign Recognition via Deep Learning on Tightly-coupled Processor Arrays at Hannover Messe 2022: The Chair for Hardware/Software Co-Design demonstrated in cooperation with the FAU Research Center Embedded Systems Initiative (FAU ESI) a Hand Sign Recognition via Deep Learning on Tightly-coupled Processor Arrays (TCPAs) as part of the joint booth of BayernInnovativ at Hannover Messe 2022. The interest in „Future Hub“ (Hall 2) was unbroken despite the smaller edition of Hannover-Messe. The number of visitors at our booth was comparable with former years. We are happy about the many exciting conversations we had.
Convolutional Neural Networks (CNNs) are the method of choice für implementing deep-learning strategies on high-end accelerators (i.e. GPUs or FPGAs). Embedded Systems lack however computing power needed for theses implementations. Tightly Coupled Processor Arrays are ideal for energy-efficient acceleration of programs containing nested loops. The demonstrator shows a hand sign recognition, which is accelerated on a TCPA. The movement of a person’s hand is captured by a webcam. The demonstrator then tries to imitate the shown gesture in real-time using a robotic hand. Therefore, the video is sent to an FPGA, on which the TCPA executing a trained CNN is implemented as a prototype. This results in an immense speedup compared an execution of the CNN on a CPU desire its much lower clock rate. Current work is on implementing the TCPA on a chip using 22 nm technology.