Case Study
Criminology images analysisThrough Computer vision methodology, our client wants to locate possible DNA areas on images coming from crime scenes to accelerate the process of analysis.
IN BRIEF
KEY INFORMATION
DNA LOCATION
Bounding box using deep learning
AUTO-LABEL
With an intelligent use of model’s probabilities
GLOBAL-LEARNING
Generalization to items never encountered during the training phase
Challenges
- Generalize on objects the algorithm doesn’t know thanks to a future-proof model
- Consider the laboratory confidence scores for heatmaps visualization
- Create an autonomous solution which could be deployed on an isolated machine
Solution
- Creation of an automatized labelling solution for possible DNA areas on images
- Rectangles of localization & probabilistic heatmaps for results.
- Time saving from the photo shooting to the sample
Gains
- Pictures analyzed: 2000
- History of all analysis stored through time and easy to browse
- Ability to analyze any kind of objects found on crime scene
Testimonial
“This solution has completely changed our way of working and we
no longer waste time”