Path construction method and apparatus, terminal, and storage medium

A path construction method, applied to an automatic driving vehicle, comprising:

• when the vehicle is traveling on a driving path of an underground garage at a first time, acquiring a first vehicle travel state information and a first initial image of a surrounding environment of the driving path corresponding to the first vehicle travel state information in real time, the underground garage includes a plurality of travelable paths, the driving path is one of the travelable paths, wherein the first vehicle travel state information comprises a travel strategy when the vehicle is traveling on the driving path and a driving habit of a driver;
• calculating a first target top view corresponding to the first initial image via a nonlinear difference correction algorithm according to the first initial image;
• inputting the first target top view into a preset deep learning model, and classifying pixel points of the first target top view input into the preset deep learning model to obtain a first partitioned image, the first partitioned image comprising a first travelable region image of a travelable region and a first non-travelable region image of a non-travelable region, a travelable region includes a travelable road and a travelable intersection, a non-travelable region includes a parking space line and a parking space area;
• scanning the first partitioned image to recognize a first travelable region of the vehicle;
• and
• generating a first path trajectory corresponding to the first vehicle travel state information based on the first travelable region;
• and
• further comprises:
• when the vehicle is traveling on the driving path at a second time, acquiring a second vehicle travel state information and a second initial image of a surrounding environment of the driving path corresponding to the second vehicle travel state information in real time;
• calculating a second target top view corresponding to the second initial image via the nonlinear difference correction algorithm according to the second initial image;
• inputting the second target top view into the preset deep learning model, and classifying pixel points of the second target top view input into the preset deep learning model to obtain a second partitioned image, the second partitioned image comprising a second travelable region image of the travelable region and a second non-travelable region image of the non-travelable region;
• scanning the second partitioned image to recognize a second travelable region of the vehicle;
• and
• generating a second path trajectory corresponding to the second vehicle travel state information based on the second travelable region;
• and
• performing multi-trajectory fusion on the second path trajectory and the first path trajectory, and
• reconstructing a map corresponding to the driving path;
• and
• wherein before performing multi-trajectory fusion on the second path trajectory and the first path trajectory, further comprising:
• judging whether a coincidence degree between the second path trajectory and the first path trajectory is greater than or equal to a preset first threshold value;
• and if the coincidence degree between the second path trajectory and the first path trajectory is greater than or equal to the preset first threshold value, fusing the second path trajectory and the first path trajectory.