INTRODUCTION OF AI FOR EXTRACTION FROM A LARGE NUMBER OF IMAGES
At ARAYA, we use image recognition technology based on deep learning to solve a variety of business issues for our customers.
Here, we introduce a case where AI was introduced to the task of extracting the best images that match the conditions for publication in the media from a large number of images taken at a sports game.
CHALLENGES BEFORE THE INTRODUCTION OF AI
In the past, the client took a large number of photos of sports games and manually extracted the best images that met the conditions such as "full body" and "competition equipment" from them.
So, we had the following issues.
1. time consuming selection
We used to check all the images visually; in a single match, the number of images taken could be several thousand, which took about three to four hours.
2. required to be extracted in a short time
The extracted images needed to be extracted in a short time in order to be published in the media. In addition, it was necessary to secure a coherent time for the selection after every game.
EFFECTS AFTER AI IMPLEMENTATION
TO SOLVE THE ABOVE ISSUES, WE WERE ABLE TO SIGNIFICANTLY REDUCE THE NUMBER OF IMAGES TO BE CHECKED BY CREATING A SYSTEM IN WHICH AI EXTRACTS IMAGES THAT MATCH THE CONDITIONS FROM AMONG THOUSANDS OF TARGET IMAGES.
Specifically, the following effects were obtained.
1. drastic reduction of selection time
It now takes only a few minutes for the AI to extract images that match the conditions from thousands of images.
2. almost no need to set aside time for extraction
Since the extraction process is performed by AI, it is no longer necessary for people to set aside several hours of their time.
POINTS TO CONSIDER WHEN DEVELOPING AI 1: COMBINATION OF EXTRACTION CONDITIONS
By combining AI algorithms and rule-based processing, we were able to extract the data that satisfied multiple conditions desired by the customer.
Specifically, we performed the following processes.
Object detection algorithm (SSD) to detect people and competition equipment.
Determine whether the whole body is in the image by estimating the key points of the person.
The following images are excluded by rule-based processing.
The following images are excluded by rule-based processing: ・The image is far away ・The tool is not near the person
POINTS TO CONSIDER WHEN DEVELOPING AI 2: SELECTING THE BEST ALGORITHM
We selected and proposed the most suitable AI algorithm to realize the customer's desired needs.
For example, for the aforementioned "determining whether the entire body is in the picture", ARAYA proposed a method called key point estimation. A key point is a human joint point, and we judged whether or not the subject's entire body was captured by counting a certain number of key points.
SUMMARY: FOR THOSE WHO ARE CONSIDERING THE INTRODUCTION OF AI TO IMPROVE THE EFFICIENCY OF OPERATIONS RELATED TO IMAGES AND VIDEO
Please consult with ARAYA regarding the introduction of AI to improve efficiency and automation of image and video related operations.
For example, if you have any of the following issues regarding the introduction of AI, we can advise you after careful consideration of possible solutions.
If you are manually checking a large number of images or videos
When there are many conditions to be extracted, etc.