AI security players focus on ReID technology. How is it currently developing?

What should I do if the facial recognition camera cannot see the face? Since 2015, when facial recognition surpassed human accuracy, the recognition accuracy has jumped 100,000 times in four years. In the future, it will unlock more complex application scenarios and improve efficiency by several orders of magnitude. However, in current actual application scenarios, a single camera cannot cover all areas, and the detection ranges of multiple cameras rarely overlap. Moreover, due to the installation height and density of video surveillance probes, more pedestrians’ side faces, the top of their heads, or the back of their heads are captured.

Before face recognition is realized, there is another problem: cameras are everywhere in the city. Except for the bayonet cameras and the snapshot cameras used by the electronic police, there are very few cameras that meet the face collection standards: the existing front-end equipment is difficult to meet the requirements of the installation height and angle required for face recognition, and needs to be replaced or renovated, which involves a long period of high investment. To maximize the potential of the cameras installed in the Safe City and Snow Bright Project, other technologies need to be added.

The leading players in the AI ​​+ security industry are beginning to focus on the combined application of ReID (pedestrian re-identification) technology and face recognition. ReID technology does not have high requirements for the environment and equipment, and can make full use of existing equipment and infrastructure. As long as certain pedestrian features are identified, it can achieve continuous tracking and trajectory restoration of the search object across cameras. Once facial information is found in this process, the identity of the person can be determined. This is an ideal technology for social public security.

For example, if you know a suspect's photo and want to obtain the video segment in which he appears in the surveillance, ReID can collect data from the surveillance video library based on the photo and connect the suspect's trajectory in each camera, which will be of great help to criminal investigation work.

In general, in the field of surveillance, manufacturers hope to use ReID technology to associate and find the same object in different videos. Only when this level is achieved can the entire surveillance operation be considered complete and the overall situation can be seen.

It should be pointed out that ReID itself is a very difficult technology. To identify the same person from different videos, video lighting conditions, monitoring area resolution and angle, target occlusion, and even people wearing similar clothes can all make identification difficult. The difficulty in obtaining data will pose a greater challenge to the algorithm. The industry is now trying to work on the algorithm level to improve ReID recognition results.

Established security companies such as Hikvision and Dahua, the four AI unicorns, Huawei, and traditional Internet giants such as BAT all have their own layout and accumulation in the field of ReID. CloudWalk's ReID technology has surpassed Alibaba, Tencent, and the Chinese Academy of Sciences, breaking three world records and pushing the ReID recognition rate to a commercially viable level.

Conclusion: "Face recognition" as an important application in the field of computer vision has swept the entire industry, and "recognizing people" will push the computer vision industry into a new stage of development. At present, ReID technology is mainly used in the security field. The security industry has conducted in-depth research in many sub-fields of ReID, such as pedestrian detection, posture estimation, behavior recognition, and pedestrian structured information extraction. In the future, the combination of ReID and face recognition can be applied to more and richer scenarios, allowing everyone to more quickly experience the upgrade of intelligent technology in security, autonomous driving, home life and other aspects.