Underground parking lot high-precision visual positioning method and system based on VGG + NetVLAD
An underground parking lot and visual positioning technology, applied in the field of computer vision, can solve the problems of poor generalization ability and poor robustness, and achieve the effect of fast search speed, strong robustness and high accuracy
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0074] Example 1. Visual Map of Safety Exit Signs in Underground Parking Lot
[0075] Such as figure 1 Shown, the present invention utilizes the safe exit sign in underground parking lot to make visual map, and the concrete operation that realizes this method is as follows:
[0076] 1. Data collection stage: collect the image of the safety exit sign in the target underground parking lot, and record the required data information.
[0077] Use the selected camera to shoot the signs in the parking lot, pay attention to: the collected signs cover the parking lot as much as possible; the image content includes certain scene features; the sign is complete in the image; each sign collects several images, including different lighting conditions, different Shooting distance, different shooting angles.
[0078] The data information corresponding to each sign recorded during shooting includes: the location information of the sign; the color, shape and geometric size of the sign. In t...
Embodiment 2
[0089] Embodiment 2. High-precision visual positioning method for underground parking lot
[0090] Combine below Figure 4 The algorithm flow of the positioning method further describes the present invention. In the target underground parking lot, take an image of the safety exit sign at the location, and use the prepared sign map to realize high-precision positioning of the underground parking lot, including the following steps:
[0091] 1. Types of identification marks.
[0092] Input the captured image of the underground parking lot as the query image into the YOLOv3 sign detection network model, detect the sign in the scene, identify the sign type, return the sign type and the image coordinates of the detection frame, and call the set of signs of this type in the map at this time;
[0093] 2. Search for the nearest logo image.
[0094] Input the query image into the feature extractor to extract the visual vector, and the visual vector V of the query image q Visual vect...
Embodiment 3
[0118] Embodiment 3. Positioning result of an underground parking lot
[0119] Collect sign image data in the target underground parking lot, use the method proposed by the present invention to make a visual map of the safety exit sign, and test the positioning method proposed by the present invention. The changing factors of the test image include different light intensity, different shooting distances and angles.
[0120] The present invention realizes visual positioning based on the method of deep learning (YOLOv3 and VGG+NetVLAD), and can recognize the specific identity of the input test image under different illumination, distance and angle, and the recognition accuracy rate exceeds 90%.
[0121] Then extract the feature points of the sign, and realize the pose calculation through the homography matrix, and the distance from the camera to the sign plane can be obtained. Some test error results are as follows: Figure 5 As shown, the minimum distance error is about 3mm, an...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


