Abstract: A method for recognizing a parking space for a vehicle and a parking assistance system are disclosed. An obstacle is identified from successive image frames captured when the vehicle is moving and a first boundary for the obstacle is generated by a Convolutional Neural Network (CNN) algorithm based on a position of the obstacle shown in each of the successive image frames. Distances between the moving vehicle and the obstacle are detected by ultrasonic sensors. A second boundary for the obstacle is generated by a distance modification module based on the distances between the vehicle and the obstacle. A periphery of the obstacle is defined by a periphery definition module. In view of the periphery of the obstacle, a parking space is thus recognized by a parking space recognition module. The parking process can be changed to a self-drive mode, and remotely controlled by a mobile device.