Unlock instant, AI-driven research and patent intelligence for your innovation.

Object detection method, device and system

A target detection and target object technology, applied in the field of target detection, can solve the problems of long time, slow target detection, and difficult calculation of target detection models, and achieve the effect of reducing the number of candidate frames and improving the speed of target detection.

Active Publication Date: 2018-07-10
AXERA SEMICON (SHANGHAI) CO LTD
View PDF7 Cites 41 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] However, the above-mentioned process of extracting an optimal frame from multiple candidate frames is very complicated. The non-maximum value suppression algorithm is used as an example to illustrate: first, all candidate frames are sorted by confidence, and the candidate frame with the highest confidence is selected; Then traverse the rest of the candidate boxes, if the overlapping area of ​​a candidate box and the candidate box with the highest confidence is greater than a certain threshold, delete this box, after the traversal, leave the candidate box with the highest confidence, and then do the remaining Repeat the above process until all the candidate frames are processed, and the final candidate frame is the optimal frame extracted by the non-maximum value suppression algorithm that can accurately represent the position of the target object in the image to be detected
This process of extracting the optimal frame from N candidate frames involves confidence sorting and multiple "traversal-elimination" iterative processes. It is a detection algorithm with a complexity of N*N. If the total number of candidate frames is N When the value is large, the target detection model is difficult to calculate, takes a long time, and the target detection speed is slow

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Object detection method, device and system
  • Object detection method, device and system
  • Object detection method, device and system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0032] First, refer to figure 1 An example electronic device 100 for implementing the object detection method, device and system of the embodiments of the present invention will be described.

[0033] Such as figure 1Shown is a schematic structural diagram of an electronic device. The electronic device 100 includes one or more processors 102, one or more storage devices 104, an input device 106, an output device 108, and an image acquisition device 110. These components pass through a bus system 112 and / or other forms of connection mechanisms (not shown). It should be noted that figure 1 The components and structure of the electronic device 100 shown are only exemplary, not limiting, and the electronic device may also have other components and structures as required.

[0034] The processor 102 can be implemented in at least one hardware form of a digital signal processor (DSP), a field programmable gate array (FPGA), and a programmable logic array (PLA), and the processor 1...

Embodiment 2

[0041] refer to figure 2 The flow chart of a target detection method shown specifically includes the following steps:

[0042] Step S202, acquiring a plurality of primary candidate frames corresponding to the image to be detected; wherein, the primary candidate frames are used to represent a preliminary estimated position of the target object in the image to be detected.

[0043] The primary candidate frame in this embodiment may also be called a detection window. Each primary candidate frame may contain a part of the target object, and multiple primary candidate frames of the same target object may overlap. The ultimate goal of target detection is to find the one that is most likely to completely contain the target object from multiple primary candidate frames Preferred box.

[0044] In practical applications, multiple primary candidate boxes corresponding to the image to be detected can be obtained through the convolutional neural network CNN; among them, the convolutiona...

specific Embodiment approach

[0053] Method 1: Threshold filtering

[0054] In this method, the frame information includes confidence, and according to the frame information of each primary candidate frame, the step of determining the secondary candidate frame from multiple primary candidate frames is performed as follows:

[0055] (1) The confidence of each primary candidate box is compared with a preset confidence threshold.

[0056] (2) Determine the primary candidate frame whose confidence is higher than the preset confidence threshold as the secondary candidate frame.

[0057] Assuming that the same target object (for example, a certain face) on the image to be detected corresponds to five primary candidate frames, the confidence of each primary candidate frame is 0.98, 0.75, 0.53, 0.42, and 0.32; set the preset confidence threshold is 0.5, the three primary candidate frames with confidence levels of 0.98, 0.75, and 0.53 are determined as secondary candidate frames, and the two primary candidate fram...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The present invention provides an object detection method, device and system, and relates to the technical field of object detection. The method comprises the steps of: obtaining a plurality of primary candidate frames corresponding to images to be detected, wherein the primary candidate frames are configured to show a preliminary estimation position of a target object in the images to be detected; according to frame information of each primary candidate frame, determining secondary candidate frames from the primary candidate frames, wherein the number of the secondary candidate frames is smaller than the number of the primary candidate frames; and inputting the determined secondary candidate frames into an object detection model to allow the object detection model to determine a target frame on the images to be detected according to the secondary candidate frames, wherein the target frame is configured to show a final estimation position of the target object in the images to be detected. The object detection speed can be effectively improved.

Description

technical field [0001] The present invention relates to the technical field of target detection, in particular to a target detection method, device and system. Background technique [0002] In many computer vision tasks such as target detection, many target objects such as faces, pedestrians, and cars in the image need to be detected. Usually, the neural network will first perform preliminary detection on the image to be detected to obtain multiple candidates corresponding to each target object. frame, the multiple candidate frames of the target object often partially overlap, so it is necessary to extract an optimal frame from the multiple candidate frames corresponding to each target object by a target detection model such as the non-maximum value suppression algorithm as the final Test results. [0003] However, the above-mentioned process of extracting an optimal frame from multiple candidate frames is very complicated. The non-maximum value suppression algorithm is use...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/32G06N3/04
CPCG06V10/245G06V10/255G06N3/045
Inventor 梁喆曹宇辉周舒畅
Owner AXERA SEMICON (SHANGHAI) CO LTD