Dear all, any update on this subject?

All the best.

Hi @kimhendrikse, interesting stuff! I'm currently working on a detection system for desert lions in Namibia. Not exactly the same as it will analyse images afterwards on a local machine instead of in real time, but still. For this I use MegaDetector to detect the animals and a yolov8 classifier to further classify them into lion or non-lion (just like the deepfaune project). 

If you're looking for camera trap images of wolves, LILA BC has around 5500 images classified as wolf (Canis lupus, shots taken from North America). If you want to include domestic dog, they have more than 10k images available (Canis familiaris, shots taken from various places on earth). Usually, If I need training data I run MegaDetector over these LILA BC images with a high threshold (±0.7) and square crop them. That should give you quite some training data to train a wolf/dog classifier. MegaDetector is also really good in detecting people, so you would get that for free. 

Anyway, I live in Utrecht and would love to be involved in this project. Maybe get in touch and meet? [email protected] 

Hi,

This is rakesh kalva from India. I have been working on human-elephant conflict in the state of Andhra Pradesh for the past 5 years. This is a cool idea. I have used camera traps to identify movement and individual elephants for demographic data. 

Some field based observations of using camera traps for elephants:

1. I usually place the camera traps on forest paths and many of the images we captured wernt of the complete elephant. Its just the legs and trunk. Wont that effect the detection by the software as to wether its an elephant or not?

2. In case of flash cameras , quite a lot of our cameras got damaged by elephants. So we were going for IR cameras. WIll it work for IR cameras as well?

3. Another issue is with the cameras being stolen by locals or poachers.

4. So we gave up all this and are using a simple trip technology with a switch attached to a door bell attached to a rope placed at a height of 7-8 feet. In this landscape no other animal is at this heigh so when the alarm is triggered we know its an elephant.

5. Elephants operate in a large territor, so will it be feasable to use this technology? 

But there are some interesting research questions that can be answered with this technology you are developing. Kindly let me know if i can be of any help on field.

Regards,

Rakesh Kalva

Wildlife Biologist

Hi Neil,

I am a project manager for Zoological Society of London's Thailand Programme. We work in the southern Western Forest Complex of Thailand, which is a large landscape of relatively contiguous forest surrounded by development and agriculture and thus rife with human-elephant conflict. We are currently looking to co-develop or pilot low-cost cameras or acoustic sensors for real-time detection of elephants at HEC hotspots within the landscape. Current issues we've been facing are high rates of non-target stimuli triggers which clog up the cloud (so the automated classification of elephants would be useful to limit notifications) and the high cost of conventional cellular camera systems. I am curious to know the current status of your Elephant AI system. The most recent update I've found on your hackaday is of the combination with a deterrence system, which is very promising. Feel free to email me at [email protected] or reply to this message.

Cheers,

Juliana