In the field of intelligent transport systems, driver assistance systems are of major importance. These systems are designed to display to the driver relevant info from the surrounding environment.
Abstract
In the field of intelligent transport systems, driver assistance systems are of major importance. These systems are designed to display to the driver relevant information from the surrounding environment. This information may include distance from other cars in different routes, the existence of pedestrians and more. Estimation of this data and its analysis can help the driver to better evaluate his current driving state and help avoid dangerous situations.
An image processing based system can be integrated in a variety of vehicles regardless of their manufacturer. This project is an attempt to create a night time system in Matlab environment. The system will detect vehicles and distances in the road in difficult light conditions. Identifying objects found on the road is a difficult task, more so in night time. Experiments that have been done during actual trips at night time showed that the algorithm allows a fast and reliable detection of vehicles in a variety of dark environments with different levels of lighting.
Flowchart
Basic assumptions
- The night vision camera is connected to the car
- License plate identification is impossible from large distance
- Cars will always be in specific area of the frame
- Cars movement is continuous
- The cars two back lights are always on
- The background isn’t stationary
- Car detection is only for private vehicles with two back lights
- In private vehicles the range between flash light center is a fixed value
- Cars flash light move together
- There is a noise in the picture that can look like a flash light
- Flash light is an object with pre defined geometry and color
- There are two different types frames
Problems with identifications
- Back light size is not constant
- Distance between back lights is not proportional to their size
- Street light may look like a back light
- Vehicles may bounce
- Back light auras
- Break lights
- Multiple back lights at the same distance
Algorithm image manipulations
- Median blur – using 3X3 matrix
- Black white transform using const value
- Erode using 3×3 square
- Blob analyzer – blob from 4 px to 5000 px
Results
Conclusions
- The system can identify cars with high probability.
- This system is not reliable by itself and human interference in needed.
- This system is vulnerable to rainy conditions