Obstacle classification method and apparatus based on unmanned vehicle, device, and storage medium

1. An obstacle classification method based on an unmanned vehicle, comprising:

obtaining obstacle information of a plurality of obstacles detected by a detection device of the unmanned vehicle; and

performing classification processing on the obstacle information of the plurality of obstacles using a compressed random forest model, to obtain a result of obstacle classification, wherein the compressed random forest model comprises at least two decision trees; a non-leaf node of each decision tree of the at least two decision trees stores only feature index information, classification threshold index information and node position index information, wherein the node position index information is left node position index information or right node position index information, the left node position index information is used to indicate positions of a left node and a right node under the non-leaf node, and the right node position index information is used to indicate the positions of the left node and the right node under the non-leaf node; a leaf node of each decision tree stores only class index information,

wherein the performing classification processing on the obstacle information of the plurality of obstacles using a compressed random forest model, to obtain a result of obstacle classification comprises:

distributing the obstacle information of the plurality of obstacles into the decision trees of the compressed random forest model;

performing, in each decision tree of the compressed random forest model, classification processing on the obstacle information inputted into the each decision tree, to obtain obstacle class information outputted from the leaf node of each decision tree; and

obtaining the result of the obstacle classification according to the obstacle class information outputted from the leaf node of each decision tree,

wherein the performing, in each decision tree of the compressed random forest model, classification processing on the obstacle information inputted into the each decision tree, to obtain obstacle class information outputted from the leaf node of each decision tree comprises:

performing, in each decision tree of the compressed random forest model, the classification processing on the obstacle information inputted into the each decision tree, and determining whether or not the non-leaf node of each decision tree has the left node or the right node;

storing the feature index information, the classification threshold index information, and the right node position index information in the non-leaf node of each decision tree in response to the non-leaf node of the each decision tree having the right node but not having the left node;

storing the feature index information, the classification threshold index information, and the left node position index information in the non-leaf node of each decision tree in response to the non-leaf node of the each decision tree having the left node but not having the right node;

storing the feature index information and the classification threshold index information in the non-leaf node of each decision tree, and store the left node position index information or the right node position index information in the non-leaf node of each decision tree, in response to the non-leaf node of the each decision tree having both the left node and the right node; and

obtaining, according to a process of the classification processing, the obstacle class information outputted from the leaf node of each decision tree.

2. The method according to claim 1 , wherein after the performing classification processing on the obstacle information of the plurality of obstacles using a compressed random forest model, to obtain a result of obstacle classification, the method further comprises:

storing information that is stored in the leaf node of each decision tree of the compressed random forest model into a parent node corresponding to the leaf node; and

deleting the leaf node of each decision tree of the compressed random forest model.

3. The method according to claim 1 , wherein the non-leaf node comprises a root node, a node having a child node and a parent node.

4. The method according to claim 1 , wherein the left node position index information is information on a relative address of the left node of each node in the decision tree relative to the right node of the each node;

the right node position index information is information on a relative address of the right node of each node in the decision tree relative to the left node of the each node.

5. The method according to claim 1 , wherein a field of the feature index information occupies 2 bytes, a field of the class index information occupies 2 bytes, a field of the classification threshold index information occupies 4 bytes, a field of the left node position index information occupies 4 bytes, and a field of the right node position index information occupies 4 bytes.

6. The method according to claim 1 , wherein the detection device is any one of the following:

a laser radar sensor, ultrasonic radar, an image detector, and an infrared detection device.

7. The method according to claim 1 , wherein the obstacle information comprises at least one of the following:

a movement speed of an obstacle, a volume of the obstacle, heat of the obstacle, and a distance between the obstacle and the unmanned vehicle.

8. An obstacle classification apparatus based on an unmanned vehicle, comprising a transmitter, a receiver, a memory and a processor;

the memory is configured to store computer instructions; the processor, when executing the computer instructions stored in the memory, is configured to:

obtain obstacle information of a plurality of obstacles detected by a detection device of the unmanned vehicle; and

perform classification processing on the obstacle information of the plurality of obstacles using a compressed random forest model, to obtain a result of obstacle classification, wherein the compressed random forest model comprises at least two decision trees; a non-leaf node of each decision tree of the at least two decision trees stores only feature index information, classification threshold index information and node position index information, wherein the node position index information is left node position index information or right node position index information, the left node position index information is used to indicate positions of a left node and a right node under the non-leaf node, and the right node position index information is used to indicate the positions of the left node and the right node under the non-leaf node; a leaf node of each decision tree stores only class index information;

wherein the processor is further configured to:

distribute the obstacle information of the plurality of obstacles into the decision trees of the compressed random forest model;

perform, in each decision tree of the compressed random forest model, classification processing on the obstacle information inputted into the each decision tree, to obtain obstacle class information outputted from the leaf node of each decision tree; and

obtain the result of the obstacle classification according to the obstacle class information outputted from the leaf node of each decision tree;

wherein the processor is further configured to:

perform, in each decision tree of the compressed random forest model, classification processing on the obstacle information inputted into the each decision tree, and determine whether or not the non-leaf node of each decision tree has the left node or the right node;

store the feature index information, the classification threshold index information and the right node position index information in the non-leaf node of each decision tree if it is determined that the non-leaf node of the each decision tree has the right node but does not have the left node;

store the feature index information, the classification threshold index information and the left node position index information in the non-leaf node of each decision tree if it is determined that the non-leaf node of the each decision tree has the left node but does not have the right node;

store the feature index information and the classification threshold index information in the non-leaf node of each decision tree, and store the left node position index information or the right node position index information in the non-leaf node of each decision tree, if it is determined that the non-leaf node of the each decision tree has both the left node and the right node; and

obtain, according to a process of the classification processing, the obstacle class information outputted from the leaf node of each decision tree.

9. The apparatus according to claim 8 , wherein the processor is further configured to:

store information that is stored in the leaf node of each decision tree of the compressed random forest model into a parent node corresponding to the leaf node after performing the classification processing on the obstacle information of the plurality of obstacles using the compressed random forest model and obtaining the result of the obstacle classification; and

delete the leaf node of each decision tree of the compressed random forest model.

10. The apparatus according to claim 8 , wherein the non-leaf node comprises a root node, a node having a child node and a parent node.

11. The apparatus according to claim 8 , wherein the left node position index information is information on a relative address of the left node of each node in the decision tree relative to the right node of the each node;

the right node position index information is information on a relative address of the right node of each node in the decision tree relative to the left node of the each node.

12. The apparatus according to claim 8 , wherein a field of the feature index information occupies 2 bytes, a field of the class index information occupies 2 bytes, a field of the classification threshold index information occupies 4 bytes, a field of the left node position index information occupies 4 bytes, and a field of the right node position index information occupies 4 bytes.

13. The apparatus according to claim 8 , wherein the detection device is any one of the following:

a laser radar sensor, ultrasonic radar, an image detector, and an infrared detection device.

14. The apparatus according to claim 8 , wherein the obstacle information comprises at least one of the following:

a movement speed of an obstacle, a volume of the obstacle, heat of the obstacle, and a distance between the obstacle and the unmanned vehicle.

15. A non-transitory computer-readable storage medium, comprising a readable storage medium and computer instructions stored in the readable storage medium; the computer instructions are used to implement the obstacle classification method based on the unmanned vehicle according to claim 1 .