Automobile drivable area planning method based on multi-task neural network

Abstract

The invention provides an automobile drivable area planning method based on a multi-task neural network, and relates to the fields of deep learning, computer vision, auxiliary driving, image processing and the like. The method comprises the following steps: firstly, constructing a lightweight multi-task neural network with two functions of semantic segmentation and target detection based on a hardparameter sharing mechanism; secondly, according to a network output format, making a training set and constructing a corresponding loss function mathematical model; training the network again, and performing back propagation by using the loss function mathematical model provided by the invention to optimize network parameters; finally, applying the multi-task neural network to automobile drivable area planning. The multi-task neural network not only has a lane segmentation function, but also has a vehicle and pedestrian detection function, an automobile anti-collision early warning functioncan be achieved through monocular distance measurement, and the probability of automobile collision is reduced.

Description

technical field [0001] The invention belongs to the fields of deep learning, computer vision, assisted driving, image processing and the like, and in particular relates to a multi-task neural network applied to the planning of a vehicle's drivable area. Background technique [0002] With the development of deep learning, autonomous driving technology is becoming more and more mature. Some companies have produced partially autonomous vehicles and tested them on the road. However, autonomous driving still has some problems in the field of environmental perception. Environmental perception in autonomous driving requires segmenting lanes and judging which lane is a drivable lane, as well as detecting targets such as vehicles and pedestrians ahead to prevent collisions. Segmenting lanes and detecting objects belong to two different functions, namely semantic segmentation and object detection. Running semantic segmentation and object detection neural network in the vehicle chip a...

AI Technical Summary

Problems solved by technology

Simultaneously running semantic segmentation and object detection neural networks in the on-board chip will consume a lot of computing resources on the on-board chip and affect other functions

[0003] To sum up, the problem existing in the existing technology is that multiple neural networks run simultaneously in one vehicle-mounted chip, which will generate a huge load on the chip and consume a large amount of computing resources.

[0007] First of all, the document CN1111178253A is only a general introduction to the multi-task neural network applied to automatic driving. Some of the steps in his document are very broad and general processes, which do not have guiding and practical significance. ; Then, what kind of network should be selected, and how to combine and connect between multiple networks has not been submitted for description; secondly, how to construct the loss function, and what loss function to choose is also not described, which is also a simple linear weighted summation; finally in the document CN1111178253A The multi-task neural network mentioned has too many task branches, and too many task branches will cause two problems: 1. It is difficult to train, and each task has its own unique attributes. If only simple linear Weighted sum loss function, personal task network cannot be trained successfully; 2. Too many parameters and many task branches will increase the number of parameters in the total network, and the amount of calculation will increase accordingly, so the calculation speed of the network will be very slow. It cannot meet the real-time requirements and has no practical value. For example, mask-rcnn, the network has only two task branches, and the detection speed of the network is relatively slow.

Images