Also wanted to mention the Diopsis camera - https://diopsis.eu/en/camera/. 

Not sure what the folks at apic.ai use? 

As already mentioned by Tom in the first post, we are using the OAK-1 (OpenCV AI Kit) camera in our insect-detecting camera trap. You can find more info about the hardware here.

• High resolution (without going overboard)

  We are using 4K frames (3840x2160 px) at the moment, which is good enough (for classification of cropped detections) for bigger insects (e.g. bees, hoverflies, wasps) at a distance of about 40cm from camera to object.

• Cheap

  There is also a cheap 99 $ version with lower quality camera, which might be good enough depending on the project.

• Low power

  For our processing pipeline the total power consumption (including the RPi Zero 2 W) is about 3.7 W (740mA @ 5V).

• Good interface for a Raspberry Pi or similar

  Very easy to use Python API.

• Good white balance

  Just from looking at the images the white balance seems good, but I can't say anything more detailed about this point.

• Reliable auto focus

  Auto focus is very quick, the camera even focuses on insects flying through the frame.

• Appropriate depth of focus/zoom for the target

  Auto-Focus: 8cm - ∞

Because this is basically a camera module directly connected to a VPU chip on-board, you can run lightweight detection (and other) models on-device. This means that there is (almost) no load on the host device, which makes it possible to use the camera in combination with e.g. a Raspberry Pi Zero. All work is done on the camera chip, while the Raspberry only has to send the processing pipeline (Python script) to the camera and receive the processed data (e.g. detections, raw/cropped frames etc.).

What makes this device great for development is the completely open sourced resources. You can immediately start with a off-the-shelf model and scripts available at Github.

From our experience, the OAK camera works great with lightweight detection models like YOLOv5n or YOLOv5s. In combination with a object tracker we are currently achieving 4-5 FPS for multiple objects (insects) in the same frame.

At the moment I am working on a documentation website where you will find information about how to build our hardware setup for your own project and many details on different Python scripts to e.g. run object detection and save detected insects (or other animals!) with associated tracking IDs. We will also include a simple workflow to train detection models with your own data.

If anyone is interested in more details regarding the OAK cameras please don't hesitate to write me a message! :)

No - the trap was in their path and they just walked through it. I've now moved it to a place they can't go. The biggest threat to the moths is from pied currawongs. I schedule the trap so it shuts off at least two before sunrise to try to avoid them feasting on the larger insects.

At first I was finding wings below the screen in the morning when I put our units out. So I put a game camera on the units to see what was feeding and when. I found three bird species, likely 3 individuals, quickly found it to be a good bird feeder- Song Sparrow (most frequent), House Wren, and this Tufted Titmouse. I changed my units to turn off about 1.5 hours before dawn and that worked! Nearly all the moths left the scene before the birds came to visit. 

My most prized camera trap image - a hummingbird caught on camera!

I'm not part of the dev team and not involve in the project but this thread remind me the critterpic camera trap for cryptic species. it's design to provides real-time monitoring of small cryptic species and pest species through instant image feeds and automatic species identification. But it could also work for some type of insect.

InsectDetect camera trap

As part of the project MonViA, we are searching for alternative (continuous, non-invasive) monitoring methods to assess insect diversity and abundance in agricultural sites.

Since last year, we are developing a camera trap for automated monitoring of flower-visiting insects, based on the Luxonis OAK-1 in combination with the Raspberry Pi Zero 2 W.

A specifically trained YOLOv5n detection model running on the camera chip detects insects landing on a platform with artificial flowers in real-time. To avoid repeated counting of an individual insect in the same frame, an object tracker is used to track insects moving on the platform. For each detection, a crop of the insect is saved, combined with relevant information like confidence score, tracking ID, bounding box dimensions and exact recording time. This data is ready to be used to train classification models or for inference with available models (e.g. iNaturalist). Due to its low power consumption, a small 9 W solar panel is sufficient to power the device. Two lithium-ion batteries as backup power source and the PiJuice Zero pHAT for precise power control make the device truly self-sufficient and enable continuous monitoring during a whole season.

Assembling and programming the camera trap is realtively simple, even for beginners. All Python scripts are based on open source resources available on GitHub. A dedicated website with detailed instructions for Hardware assembly, model training and programming will be available soon. Overall cost for the device is between 500-700 €, depending on the additional mounting fixtures and hardware components. With this project we want to enable especially Citizen Scientists to build their own smart camera traps for monitoring insects on meadow orchards (or their backyards).

Hi all,

I am Chantal Huijbers, working at Naturalis Biodiversity Center in the Netherlands. I have a background in biology, but am now working as a team lead in the ARISE program to develop a national infrastructure for biodiversity monitoring and digital species identification. One of my main goals is to translate the needs of the research community to IT specialists, so that we can develop the tools to make data better available to the wider research community. 

Besides my work for ARISE, I am also the project lead for the Diopsis insect camera project, which was established to monitor insect populations in the Netherlands. Check out my post in the Project Introductions to read more about this!

hi 

I am David Roy, working at the UK Centre for Ecology & Hydrology.  I am an ecologist and most of my career has been working on biodiversity monitoring schemes, often using citizen science approaches.  I am the head of the Biological Records Centre which has supported wildlife recording in the UK for almost 60 years.

I am particularly interested in lepidoptera and am excited by the potential of automated camera systems to better monitor their status around the world. I am one of the leads of the Easy RIDER project. 

I look forward to learning a lot from this group!

Hello everyone,

I'm Max, a PhD student working at the Julius Kühn Institute (Institute for Biological Control) in southwest Germany. During the last few years I caught insects with different traditional sampling methods like yellow pan traps and emergence traps to assess the indicator potential of parasitoid Hymenoptera in agricultural grasslands. This research is part of the project MonViA, where we also want to develop and test innovative monitoring methods that can be used for a more continuous monitoring (and don't kill the insects).

In this context, we are currently developing a camera trap with automated on-device insect detection and tracking to monitor flower-visiting insects. You can find more information about the device in this post. Our camera trap is based on low-cost hardware with low energy consumption and can be powered continuously by a battery and a 9 W solar panel. The software is based on open source Python scripts and relatively easy to program even without a lot of programming experience.

I'm really excited to discuss everything insect monitoring-related (automated or not) and would be happy if you get in touch with me, if you are interested in our camera trap setup!