Object localization is vital in computer vision to solve object detection or classification problems. Typically, this task is performed on expensive GPU devices, but edge computing is gaining importance in real-time applications. In this work, we propose a real-time implementation for unsupervised object localization using a low-power device for airport video surveillance. We automatically find regions of objects in video using a region proposal network (RPN) together with an optical flow region proposal (OFRP) based on optical flow maps between frames. In addition, we study the deployment of our solution on an embedded architecture, i.e. a Jetson AGX Xavier, using simultaneously CPU, GPU and specific hardware accelerators. Also, three different data representations (FP32, FP16 and INT8) are employed for the RPN. Obtained results show that optimizations can improve up to 4.1×
energy consumption and 2.2× execution time while maintaining good accuracy with respect to the baseline model.
Buscar en Dimension