Object detection is an important task in the field of computer vision, which aims to accurately detect and locate specific target objects from images or videos.
1. Definition of target detection problem
Object detection is the process of finding and classifying a variable number of objects in a picture .
2. Common problems in the target detection process
- Target type and quantity issues
- target scale problem
- External environmental interference problems
3. The difference between target detection and image classification
Target Detection:
Target detection not only provides a rectangular frame (that is, the location of the target detection object) , but also classifies the objects in the rectangular frame . Different colors of rectangular frames represent different categories , and provides the detection object belongs to the target category. confidence level .
Image classification:
Image classification mainly uses images as input, and the probability distributions of images belonging to different categories as output. It mainly determines the category of images. Image classification is only a part of target detection.
Summarize:
Whether it is image classification or target detection , when using deep learning technology for processing, a feature extraction step is required . For classic machine learning methods, feature extraction is usually completed by designing manual features , while deep learning often uses volumes Convolutional neural network is used to extract features.
4. Target detection VS target segmentation
Target Detection:
Mainly to locate the position of the target, the position information is usually represented as a rectangle, and the rectangle can be represented by four-dimensional data.
Target segmentation:
Each pixel needs to be divided into different categories , and the segmentation result needs to be consistent with the original image size. The output result of the same size as the original image is often obtained through upsampling or deconvolution.
- Image classification: Just specify the category to which the corresponding target belongs
- Target detection: Need to locate the target and classify it
- Target segmentation: You need to find the area occupied by the current target. The figure c above is semantic segmentation and d is instance segmentation.
- Semantic segmentation: only need to find the area occupied by the same type of objects
- Instance segmentation: not only targets at different semantic levels must be distinguished, but targets of the same category also need to be divided into different instances.
In the development process of target detection algorithms, traditional algorithms based on manual features were once the mainstream. These traditional algorithms identify target objects by designing and extracting manually designed features, including Haar features, HOG features, SIFT features, etc.
This article will deeply explore the traditional algorithm based on manual features in the target detection algorithm and introduce its principles, advantages and disadvantages , and its application in computer vision.
1. Definition of the traditional target detection algorithm based on manual features The
traditional target detection algorithm based on manual features is an early type of target detection algorithm that identifies target objects by manually designing and extracting features. These features are usually based on local information of the image, such as edges, texture, color, etc. Based on feature extraction, traditional algorithms usually use classifiers or detectors to determine whether there is a target object in the image and give the target's location and size.