AI Camera System

Since video data is quite large compared to the sensor data, it is almost impossible to send video from multiple cameras to the server and perform real-time processing from an outdoor facility where the Wi-Fi environment is not ready.Therefore, we are developing an AI camera system using edge computing and deep learning technologies, which realizea real-time image processing on site without sending the video data to the server.

Enemy Detection Camera

Multiple network cameras are installed in front of the beehive and are connected to the Raspberry Pi 4 model B+ by Wi-Fi. When hornet is detected, it is notified to the users via SNS. One Raspberry Pi can process up to 10 video data from the cameras in parallel. It is possible to cope with enemies such as mantis and birds by preparing learning data.

Bee Counting Camera

By applying an object recognition algorithm for detecting enemies, bees in front of the beehive are counted.Since overlapped bees can be detected individually, a high accuracy as counting by humans is achieved.

Beehive IR Camera

An infrared camera is used to observe the inside of the beehive.　The camera video image is encoded in mp4 format and uploaded to YouTube.

Live Video Streaming

Live streaming of network cameras installed at a beekeeping facility on the roof of Noubunkyou Building in Akasaka.

---