Target detection and identification refers to finding targets from a scene (picture), including two processes of detection (where) and identification (what).The difficulty of the task lies in the extraction and identification of candidate regions to be detected. Therefore, the broad framework of the task is as follows:
Firstly, a model for extracting candidate regions from the scene is established and then identify the classification model of the candidate region Finally fine-tune the parameters of the classification model and the location of valid candidate boxes Target detection and recognition has a wide range of applications in many fields of life.
Dynamic Video Target Detection and Tracking Technology
The traditional TV monitoring technology can only achieve the function of "clairvoyance", transmitting remote target images (original data) to the monitoring center, and the monitoring personnel can make judgment on the scene according to the visual video images.The purpose of intelligent video surveillance is to convert the original video data into a sufficient amount of "useful information" for the monitoring personnel to make decisions, so that the monitoring personnel can understand the events in a timely and comprehensive manner: "
Recently, I saw an article that counted the most frequently cited in-depth study papers from 2010 to 2016. During the postgraduate period, the direction was object detection, so I intercepted some papers related to the field.Personally, I think that object detection can be divided into many detailed studies. It is a good job to find the pain points and propose the available solutions for any detail, including: the understanding of convolution neural network model (theory/network structure);Behavioral analysis of structural parameters (optimization/regularization);Performance in different data distributions.
"Summary of Facial Expression Recognition Based on Deep Learning and Traditional Machine Learning" The existing face recognition technology is limited to the traditional machine learning algorithm. Under the conditions of light intensity, occlusion, posture transformation, etc., the traditional machine learning algorithm has poor robustness and is difficult to be applied to real life.
In the late 1970s, Suwa and others marked the continuous sequence of face images into 20 feature points, and realized the recognition of facial expressions through the comparison of these feature points.
Moving Target Tracking is an indispensable link in video surveillance system.In specific scenes, there are some classical algorithms that can achieve better target tracking effect.This paper introduces general target tracking algorithms, compares several commonly used algorithms, and introduces particle filter algorithm and contour-based target tracking algorithm in detail.Finally, it briefly introduces the processing of target occlusion, multi-camera target tracking and target tracking under camera motion. 1. General Target Tracking AlgorithmGenerally,
General Classification of Mainstream Algorithms for Moving Target Tracking is mainly based on two ideas:
a) Without relying on prior knowledge, moving targets are directly detected from image sequences, and target recognition is carried out to finally track the moving targets of interest;B) depending on the prior knowledge of the target, firstly modeling the moving target, and then finding the matching moving target in real time in the image sequence.
This article is the blogger's personal opinion and summary in the process of learning target tracking. Welcome to learn from and exchange. If there are any errors, please leave a message to point out.
In the process of learning, I have been thinking about the essence of "tracking" is What?
In "New Ideas of Object Tracking [VALSE Advanced Technology Selection and Introduction 16-18]", Dr. Wang Naiyan proposed that the essence of tracking should be verification, rather than the current mainstream tracking by detection/classification.