Robotic tracking system using a deep learning approach, implemented in Keras with a Tensorflow backend. Semester project supervised by Yuejiang Liu and Prof. Alexandre Alahi at the visual intelligence for transportation lab of EPFL.
The client is the base of the communication between the robot and the local computer. The communication is based on TCP socket. This file depends on the files final_detector.py and kalman.py . It receives a image from the Loomo robot, detects the object of interest, compute a correction of the position based on a kalman filter, using the position in the past of the object and sends back the position of detected object to the robot.
This file implements a simple kalman filter using the OpenCV library.
The detector itself is implemented in this file. It is implemented as class. When a first detector object is created, the init function loads the pre-trained weights obtained after training. The forward function allows to do a prediction based on the image given as argument. A threshold can also be chosen in order to tune the sensitivity of the detector.