Obstacle identification method, vehicle-mounted device and storage medium

A vehicle-mounted device comprising:

• a storage device;
• at least one processor;
• and
• the storage device storing one or more programs, which when executed by the at least one processor, cause the at least one processor to:
• collect radar information and original image information, and obtain fused image information by fusing the radar information with the original image information;
• acquire obstacle point cloud and non-obstacle point cloud from the fused image information, comprising:
• identifying a target object from the original image information by using a preset deep neural network;
• determining pixel points corresponding to the target object and coordinates of the pixel points;
• determining whether the pixel points corresponding to the target object are obstacle pixel points or non-obstacle pixel points;
• in response that the pixel points corresponding to the target object are obstacle pixel points, determining radar point cloud corresponding to the obstacle pixel points in the fused image information to be the obstacle point cloud;
• and in response that the pixel points corresponding to the target object are non-obstacle pixel points, determining radar point cloud corresponding to the non-obstacle pixel points in the fused image information to be the non-obstacle point cloud;
• filter out the non-obstacle point cloud from the fused image information;
• obtain at least one radar point cloud group by grouping the obstacle point cloud, comprising:
• dividing the obstacle point cloud into a plurality of sets;
• and obtaining the at least one radar point cloud group by clustering the obstacle point cloud corresponding to each set of the plurality of sets;
• and
• identify an obstacle corresponding to each group of the at least one radar point cloud group.

A non-transitory storage medium having instructions stored thereon, when the instructions are executed by a processor of a vehicle-mounted device, the processor is caused to perform an obstacle identification method, wherein the method comprises:

• collecting radar information and original image information, and obtaining fused image information by fusing the radar information with the original image information;
• acquiring obstacle point cloud and non-obstacle point cloud from the fused image information, comprising:
• identifying a target object from the original image information by using a preset deep neural network;
• determining pixel points corresponding to the target object and coordinates of the pixel points;
• determining whether the pixel points corresponding to the target object are obstacle pixel points or non-obstacle pixel points;
• in response that the pixel points corresponding to the target object are obstacle pixel points, determining radar point cloud corresponding to the obstacle pixel points in the fused image information to be the obstacle point cloud;
• and in response that the pixel points corresponding to the target object are non-obstacle pixel points, determining radar point cloud corresponding to the non-obstacle pixel points in the fused image information to be the non-obstacle point cloud;
• filtering out the non-obstacle point cloud from the fused image information;
• obtaining at least one radar point cloud group by grouping the obstacle point cloud, comprising:
• dividing the obstacle point cloud into a plurality of sets;
• and obtaining the at least one radar point cloud group by clustering the obstacle point cloud corresponding to each set of the plurality of sets;
• and
• identifying an obstacle corresponding to each group of the at least one radar point cloud group.