CAMERA TRAP IMAGE ANALYSIS

By: Lorna Mugambi, Gabriel Kiarie, Jason Kabi, Teofilo Ligawa, Victor Ruto, Lians Wanjiku and Yuri Njathi

Background

Camera traps have been used for a long time to capture images of animals in their natural habitat in a less, non-intrusive manner. These data collected can be used to carry out ecological research to increase understanding of species and habitat restoration practices.

At DSAIL, we have designed, developed and deployed 4 camera traps in the Dedan Kimathi Wildlife Conservancy to capture images of the mammalian species found there. The camera traps are based on low-cost hardware; the Raspberry Pi 2, the Raspberry Pi Zero and the OpenMV Cam H7 modules.

Internal and external parts of Dsail Camera Trap
Internal and external parts of Dsail Camera Trap

Sample Detections

Animals captured by Dsail Camera Trap
Animals captured by Dsail Camera Trap
Animal Detection
Animal Detection
Animal Classification
Animal Classification

Next Steps

We are working on getting more data from the conservancy while deploying in new locations. We’re happy to announce that we have partnered with Mugie Conservancy and are currently in the process of analyzing their camera trap data.

Links

Dataset and annotations: https://data.mendeley.com/datasets/6mhrhn7rxc/6

Raspberry Pi software: https://github.com/DeKUT-DSAIL/cameratrap-pi.git

OpenMV Cam software: https://github.com/DeKUT-DSAIL/cameratrap-openmv.git

Animal Classification Tutorial: https://github.com/DeKUT-DSAIL/dsa2022-Arusha/tree/main/camera-trap

Animal Detection Tutorial: https://github.com/DeKUT-DSAIL/ieee-africon-2023/tree/main/ml-development

Publications