Deep learning based self-driving car, deep learning based self-driving control device, and deep learning based self-driving control method
The present embodiments relate to a deep learning-based self-driving vehicle, a deep learning-based self-driving control device, and a deep learning-based self-driving control method, and more particularly, to a deep learning-based self-driving vehicle, a deep learning-based self-driving control device, and a deep learning-based self-driving control method which are capable of reliably performing self-driving control to a necessary degree in a necessary situation by accurately distinguishing between and recognizing a control target object referenced while the self-driving vehicle is traveling and a structure not referenced.
1. A deep learning-based self-driving control device comprising:
a deep learning algorithm execution unit configured to output a result of executing a deep learning algorithm on a monitoring signal obtained by monitoring a periphery of a vehicle;
a self-driving control unit configured to, when the vehicle is self-driving, control the self-driving of the vehicle on the basis of the result of executing the deep learning algorithm; and
a deep learning processing unit configured to change the deep learning algorithm with reference to driver driving information,
wherein the monitoring signal is input to an input layer of a deep neural network and the result of executing the deep learning algorithm corresponds to values of a plurality of output nodes of an output layer of the deep neural network,
wherein the deep neural network includes the input layer having a plurality of input nodes, the output layer, a hidden layer having a plurality of connection nodes for connecting the plurality of input nodes to the plurality of output nodes, and a plurality of weight lines for connecting the plurality of input nodes to the plurality of connection nodes and for connecting the plurality of connection nodes to the plurality of output nodes.
2. The deep learning-based self-driving control device of claim 1 , wherein the self-driving control unit determines whether acceleration/deceleration control for the vehicle is necessary by using one or more of an amount of acceleration/deceleration control and presence or absence of acceleration/deceleration control, which correspond to the result of executing the deep learning algorithm.
3. The deep learning-based self-driving control device of claim 2 , wherein, when it is determined that the acceleration/deceleration control for the vehicle is necessary, the self-driving control unit sets the amount of acceleration/deceleration control as a target amount of control of the vehicle to control acceleration/deceleration of the vehicle.
4. The deep learning-based self-driving control device of claim 1 , wherein the deep learning processing unit calculates an error value using a difference between an amount of calculation control corresponding to the result of executing the deep learning algorithm and an amount of driver input control included in the driver driving information and adjusts weights in the deep learning algorithm in order to reduce the calculated error value.
5. The deep learning-based self-driving control device of claim 4 , wherein, when the vehicle is not self-driving, the deep learning processing unit uses an amount of acceleration/deceleration control calculated according to the result of executing the deep learning algorithm as the amount of calculation control and uses an amount of driver manipulation acceleration/deceleration as the amount of driver input control to calculate the error value.
6. The deep learning-based self-driving control device of claim 1 , wherein through the execution of the deep learning algorithm on the monitoring signal, the deep learning algorithm execution unit determines whether an object in front of the vehicle is a control target object for self-driving control on the basis of the monitoring signal and outputs the result of executing the deep learning algorithm, which includes one or more of presence or absence of control, an amount of calculation control, and a control target object recognition result.
7. The deep learning-based self-driving control device of claim 1 , wherein the deep learning processing unit updates a weight of a weight line corresponding to the output layer and then updates a weight of a weight line corresponding to the hidden layer, in proportion to an error value calculated using a difference between an amount of calculation control corresponding to the result of executing the deep learning algorithm and an amount of driver input control included in the driver driving information, to adjust the weights in the deep neural network.
8. The deep learning-based self-driving control device of claim 1 , wherein the monitoring signal includes at least one of a radar signal and a lidar signal.
9. A deep learning-based self-driving vehicle comprising:
a sensor configured to monitor a periphery of the vehicle and output a monitoring signal;
a deep learning-based self-driving control device configured to perform a deep learning algorithm on the monitoring signal and, when the vehicle is self-driving, output a control signal for controlling the self-driving of the vehicle on the basis of a result of executing the deep learning algorithm and change the deep learning algorithm with reference to driver driving information; and
a driving device configured to drive the vehicle according to the control signal,
wherein the monitoring signal is input to an input layer of a deep neural network and the result of executing the deep learning algorithm corresponds to values of a plurality of output nodes of an output layer of the deep neural network,
wherein the deep neural network includes the input layer having a plurality of input nodes, the output layer, a hidden layer having a plurality of connection nodes for connecting the plurality of input nodes to the plurality of output nodes, and a plurality of weight lines for connecting the plurality of input nodes to the plurality of connection nodes and for connecting the plurality of connection nodes to the plurality of output nodes.
10. The deep learning-based self-driving vehicle of claim 9 , wherein through the execution of the deep learning algorithm on the monitoring signal, the deep learning-based self-driving control device determines whether an object in front of the vehicle is a control target object for self-driving control on the basis of the monitoring signal and outputs the result of executing the deep learning algorithm, which includes one or more of presence or absence of control, an amount of calculation control, and a control target object recognition result.
11. The deep learning-based self-driving vehicle of claim 9 , wherein, when the vehicle is not self-driving, the deep learning-based self-driving control device performs a deep learning training process including comparing an amount of calculation control included in the result of executing the deep learning algorithm to an amount of driver input control included in the driver driving information, calculating an error value according to a result of the comparison, and adjusting weights in the deep neural network associated with the execution of the deep learning algorithm on the basis of the error value.
12. The deep learning-based self-driving vehicle of claim 11 , wherein the deep learning-based self-driving control device stores data regarding the deep neural network, and
wherein the deep learning-based self-driving control device updates a weight of a weight line corresponding to the output layer and then updates a weight of a weight line corresponding to the hidden layer in proportion to the error value to adjust the weights in the deep neural network.
13. A deep learning-based self-driving control method comprising:
outputting a result of executing a deep learning algorithm on a monitoring signal obtained by monitoring a periphery of a vehicle;
when the vehicle is self-driving, controlling the self-driving of the vehicle according to the result of executing the deep learning algorithm; and
changing the deep learning algorithm with reference to driver driving information,
wherein the monitoring signal is input to an input layer of a deep neural network and the result of executing the deep learning algorithm corresponds to values of a plurality of output nodes of an output layer of the deep neural network,
wherein the deep neural network includes the input layer having a plurality of input nodes, the output layer, a hidden layer having a plurality of connection nodes for connecting the plurality of input nodes to the plurality of output nodes, and a plurality of weight lines for connecting the plurality of input nodes to the plurality of connection nodes and for connecting the plurality of connection nodes to the plurality of output nodes.
14. The deep learning-based self-driving control method of claim 13 , wherein the changing the deep learning algorithm comprises:
calculating an error value using a difference between an amount of acceleration/deceleration control corresponding to the result of executing the deep learning algorithm and an amount of driver manipulation acceleration/deceleration included in the driver driving information; and
adjusting weights in the deep learning algorithm in order to reduce the calculated error value.