Machine learning is more reliable in case of conventional lane-based traffic though there is heterogeneous traffic. This research is intended to use video data. If machine learning is applied in case of non-lane-based heterogenous traffic, what the challenges are there.
Hi,
The saying, A problem well-defined is half-solved seems well-applicable in this context. Clear definitions and formulations of the problem will go a long way. Depending on the specific problem you'd try to address, the cost of relaxing the assumption (about lanes) varies.
For example, if you solve a problem such as identifying car makes/types, your algorithm may not even make much use lane-specific information. However, if you are solving a problem such as estimating real-time lateral separation, then lane information can be vital.
In any case, the relaxation will bite back the approach only if your earlier ML algorithm uses lane-specific information.
Hope that helps.
I believe the main limitation is the data you can obtain from such a system. If you label every car, sensors, surrounding area, then the application scenario is appealing.
- Badges
- Science topic