Developing a Low-Cost Raspberry Pi-based Camera Trap for Wildlife Detection
A walk through the Wildlife Data
1. Data Collection:
- We deployed 4 camera traps for 7 months; from June 2021 to December 2021
- 8524 images were annotated and published.
- 8254 images from the Raspberry Pi 2 and 610 from the Pi Zero.
- 325 images from the Open MV Cam H7
- All the images are in JPG format
Sample Wildlife Images
DSAIL-Porini dataset
DSAIL Porini Animals
- Defassa waterbuck
- Sykes' monkey
- Common warthog
- Bushbuck
- Burchell's zebra
- Impala
The data annotation for the DSAIL-Porini dataset was manual with the following challenges:
- Animals overlapping in an image
- Sometimes an animal would have been captured while it's body was in the frame partly.
- Some images were too far and not visible, especially if the grass had grown too long.
- Some images were too dark
- Lens flare: this degraded the quality of some images
- Time-consuming process
- Determining the sex of animal species like the zebra, warthog and monkey was difficult.
Data Annotation Example
-
These ae other samples of images collected from the conservancy
DSAIL-Porini Sample Images
-
From the data that was collected, we performes some analysis to get a better understanding of the data.
Data Analysis
-
The data collected can be used to:
- Train machine learning models for classification and Object detection
- Train and test image-based census algorithms as it includes animal counts
- Understand animal behaviour
- Estimate the animal population growth with time
- We painted the camera traps green to camouflage them.
- We got more close-ups of animals like the waterbuck.
Camouflaging Camera Traps
- Locations where mineral supplements were provided for the animals
- Trails that led to water holes in the conservancy and areas near the waterholes
- Any other place where water would be provided by the wardens.
The decision to place the camera traps in these locations was based on:
Camera Trap Placement within the Conservancy