Object recognition and passerby counting using AI

The people in the image are whited out to protect their privacy, while the AI counts the number of pedestrians walking on the street. In addition to people, it can also recognize various objects such as cars and bicycles using machine learning. By utilizing AI to automatically detect age and gender and count the number of people 24 hours a day, 365 days a year, this data could serve as valuable market information for the AI era.

For example, if this technology is installed in a store, it can automatically count the number of passersby throughout the year. By analyzing factors such as weather, the number of passersby, and past sales data, it can predict future sales with high accuracy. These precise AI predictions contribute to optimizing store staffing and purchasing quantities, addressing SDG-related issues such as food loss, and reducing greenhouse gas emissions associated with disposal and incineration.

DX Business with AI and Digital Technology

DX is currently being promoted worldwide, with digital technologies such as AI and IoT playing a central role. We conduct joint research with companies, learning from business managers' perspectives on the challenges they face in the real world, and programming AI to increase company value and solve these problems.
In a typical company, you obtain quotations for the products you need, submit a request for approval, and then place an order once internal approval is granted. When the goods are delivered, they are processed, and when an invoice arrives the following month, payment is made via bank transfer. It is extremely time-consuming and tedious to manually check which invoice corresponds to which purchase order, as well as to carry out all these complicated processes. Additionally, there is the risk of human error. By digitizing all of this, although a human is still required for the final check, all other processes can be automated. This will address the issues of a declining birthrate and labor shortages, reduce costs such as labor and office rent, and simultaneously improve profits, productivity, and quality.

Crime detection and prevention using AI

Each store is equipped with security cameras, and by integrating AI into these systems, it is possible to automatically detect criminal activities such as shoplifting. The diagram illustrates AI-powered image recognition, where the green circular points indicate locations such as hands, elbows, and shoulders, and their coordinates can be measured in real time. Using these coordinates for motion analysis and modeling makes automatic crime detection achievable.

Monitoring camera footage 24/7 by human eyes is a challenging task. However, with AI automatically detecting incidents, triggering warning sounds, and sending notifications, the workload of store managers and security centers can be significantly reduced and automated. Additionally, AI can detect situations beyond crime, such as someone suddenly feeling unwell, crouching down, or collapsing.
In the future, we aim to integrate AI into security cameras on roads, buildings, and train stations to detect emergencies requiring medical assistance or to prevent crimes before they occur. This would involve playing a siren to deter individuals and simultaneously notifying the police or disaster prevention centers. The goal is to utilize AI to protect and safeguard human lives.

Optimal problem for automated robotic warehouse

For example, the following image shows the simulation when dealing with the optimal placement problem for an automated robotics warehouse. You can use data science and AI to find the optimal placement based on the big data of the large stock in the past.

Distribution Warehouse Optimal Placement Problem

We are working on the most suitable allocation problem of physical distribution center and a poisonous snake by a genetic algorithm (GA) in the field of the logistics and the supply chain management (SCM). The next figure is the example which solved a optimum arrangement problem of distribution centre in an actual area in GA. The left side is in the state grouped together at random first. That, ton x Kiro(baggage amount x distances) in each business office and physical distribution center repeats alternation of generations in GA, and searches the combination which becomes smallest. And the optimum arrangement result obtained finally starts to be a right figure.

Automatic crack detection of concrete by AI

Following results are examples of AI analyzing concrete images and automatically judging the presence or absence of cracks. Work in high and dangerous places can be replaced with automatic work by robots. In addition, AI will be able to automatically judge symptoms that only experts could see. The figure below is examples of concrete images. Normal is a normal image without cracks, and Abnormal is an image with cracks. Also, in parentheses is the result of AI making a decision using a machine learning model called CNN. And the numbers show the percentage of that possibility.

Solar power generation prediction by AI

We are researching the use of AI to predict the amount of power generated by solar power generation. In addition to statistical methods, we are conducting research to predict illuminance with AI based on cloud images seen from the ground and information from illuminometers.

Object Recognition/Segmentation

The upper left figure below is the input image. On the right side, an AI technology called Instance Segmentation is used to indicate the recognized person in a square frame, which is filled in according to the shape. The bottom left is the result of removing the display such as the frame and only filling it with the same method. And the lower right is the result of classifying all pixels (pixels) using Panoptic segmentation, and the ceiling, walls, roads, etc. are also identified. They can be applied to a 24-hour surveillance system in places where people should not enter. In addition, this technology can be applied to automatic driving, medical image analysis, quality inspection, etc. And in the past, it is possible to automate the work that people used to do visually.

AI evaluation of rust in equipment inspection

Rust evaluation is extremely important in equipment maintenance and risk management, and visual inspection has been performed by skilled engineers. However, due to the issue of retirement age, it has not been possible to transfer skills from experienced engineers to young engineers in time, so we are conducting joint research with companies on automatic evaluation using AI, based on a huge amount of past evaluation results and photos.

The photo above is based on the photo on the left, and the only part I want to evaluate is the metal mounting device. The image on the right uses Grad-CAM to display where the AI ​​was looking in red. Although there are many unrelated parts such as the concrete wall behind and other equipment, it can be confirmed that the AI ​​only refers to metal equipment in its evaluation.

The above figure is a mixture matrix of evaluation results using CNN and ResNET, and shows that the new model has improved AI recognition accuracy.

References

・Nameki,Toyotani,Yano et al., Interpretability and performance improvement of metal corrosion/damage detection AI model using decision tree and Grad-CAM, AI/Data Science Papers 5(3), p.303-315, November 2024
・Nameki,Toyotani et al., Comparative study of CNN and ResNet models in corrosion/damage classification of tunnel lighting fixtures, Japan Society of Directory, Association organization magazine , Vo.22, p.34-44, March 2024
・A Study on Consolidated Optimal Stock Locations for lmport and Export Freight Flows in Thailand, Sarinya Sala-ngam, Yataka Karasawa, Jun Toyotani et al., International Journal of Logistics and SCM systems, Vol.9, p.71,2016
・Toward Sustainable Operations of Supply Chain and Logistics Systems, Sala-ngam, Suzuki, Toyotani et.al., Springer, p.207, 2015
・A Case of Intermediate Treatment Facilities in Chiba Japan, Sala-ngam, Suzuki, Toyotani et.al., ICLS, Proc. The 10th Intrnational Congress on Logistics and SCM Systems(ICLS), TA21, 2015
・A Basic Research on SCM Strategy Formulation Model, Chen, Wakabayashi, Toyotani et.al.,ICLS, Proc. The 10th Intrnational Congress on Logistics and SCM Systems(ICLS), TA23, 2015
・A Case Study of the Optimization of the Location Problem and the Delivery Vehicle Routing Problem for Post Office Center in Bangkok, Sala-ngam, Toyotani et.al., ICLS, Proc. The 10th Intrnational Congress on Logistics and SCM Systems(ICLS), 2015
・Logistics Operations, Supply Chain Management and Sustainability, Sala-ngam, Suzuki, Toyotani et.al., Springer, p.525, 2014
・Optimum Position in Office of Delivering Using Guide API,Toyotani et.al.,JOURNAL OF THE JAPAN SOCIETY OF LOGISTICS SYSTEMS,11/ 1, p.91, 2011 etc.