The majority of car accidents which involve pedestrians occur due to Moving violations cause by the driver, therefor our main goal in this project was to detect
pedestrians near pedestrian crossings in low visibility conditions in order to alert drivers.
The majority of car accidents which involve pedestrians occur due to Moving violations cause by the driver, therefor our main goal in this project was to detect
pedestrians near pedestrian crossings in low visibility conditions in order to alert drivers.
Motivation and Goals
- Recognition of pedestrians crossing roads or nearby crossings in all weather and lighting conditions
- Alert nearby cars of a meaningful activity in pedestrian crossings in their path
- Robust event recognition with low false identification rate
Method
- Data collection: 2 different cameras to record pedestrian crossings:
- Thermal camera – Thermapp
- RGB Camera – Galaxy S5
- Video analysis using standard algorithms (Matlab)
- Adapt algorithms for thermal video and compare results to RGB video
Initial algorithm
Initial algorithm – RGB video results
Improved algorithm
Improved algorithm – RGB video results
Final Algorithm (frame-by-frame analysis)
Final results
Conclusions
- In general, we were able to identify pedestrians in cross overs accurately during day time in normal & thermal videos.
- During night time, thermal videos provided better results because warmer moving element is more noticeable on the background.
- The best configuration in our opinion will be to use normal camera during day time and combination of thermal & normal cams during night time/poor weather/bad lightning