Fine-grained vehicle detection method based on deep neural network

Abstract

The invention discloses a fine-grained vehicle detection method based on a deep neural network, and the method can achieve the accurate detection of the specific attitude category and contour of a vehicle through the definition output, detection and training of the network. When priori knowledge such as ground plane and camera calibration information is given, a detection result can be used for estimating a drivable area, collision time and the like, and safe driving of a driver is further assisted and guaranteed. Compared with a common target detection network, the method can output more information and can meet different application requirements. The attitude category and the contour position information of the vehicle are output, and the information is beneficial for more accurately judging the position and the driving direction of the vehicle in a road. The requirement for a sensor for data collection is low, and production and use are facilitated. Calculation of the method is completed in a common RGB image, devices such as a depth sensor or a radar are not needed, the requirement can be met only through one common camera, and the cost is low.

Description

[0001] 【Technical field】 [0002] The invention relates to a fine-grained vehicle detection method based on a deep neural network. [0003] 【Background technique】 [0004] Vehicle detection is an important task in automatic driving or assisted driving systems, which can be used to calculate the collision distance and collision time to ensure driving safety. General target detection tasks can only get rough rectangular frame detection results. The rectangular frame cannot distinguish the position of each surface of the vehicle, so it cannot accurately analyze the passable area next to the vehicle, and it is not sensitive to vehicle attitude changes. This requires the ability to detect the accurate outline of the vehicle and distinguish the side, head and tail of the vehicle to achieve fine-grained vehicle detection. [0005] There are two main methods to achieve contour detection. One is to achieve instance segmentation based on segmentation candidates or pixel classification i...

AI Technical Summary

Problems solved by technology

Although the instance segmentation method can detect the outline of the vehicle, it cannot distinguish the different faces of the vehicle, and the calculation is slow

The 3D detection method can accurately obtain the position of the vehicle, but requires a 3D sensor to collect depth information, which is expensive

Images