What about problems of non-robustness and prune to simple adversarial attacks?
Dear Walid Gomaa ,
CNN-based methods have been a tendency in the last years, especially in the field of passive-vision-based navigation (navigation that does not employ LIDAR nor radar for the identification of the road and obstacles - this will be the tendency now that autonomous cars are becoming common and you cannot pollute a downtown area with radar- and laser-scan-smog - imagine a big city 5 pm traffic jam where all cars are constantly scanning their surroundings using infrared-laser (LIDAR) - it's not feasible, pedestrian's health would be put at risk, and passive vision with, only using ambient light is the future solution).
We have produced a series of literature reviews on this area, encompassing both traditional and CNN-based solutions:
Article Passive Vision Region-Based Road Detection: A Literature Review
Technical Report Recent Trends in the Identification of Threats through the P...
Technical Report Systematic Literature Review for Passive Vision Road Obstacl...
Technical Report Systematic Literature Review for Region-based Road Detection
I hope they're useful!
Deep learning has rapidly evolved into the de-facto approach for acoustic modeling in automatic speech recognition , showing tremendous improvement in accuracy, robustness, and cross-language generalizability over conventional approaches.
CNN and RNN methods have the capability and they have proved best in the last few couple of years.
I believe so, if you check up the trendy research in this area. However, it is the only reliable technology to be used for accelerating the autonomous driving, some so-called conventional techniques such radar technologies can also play a crucial role as they also evolve significantly.
- Badges
- Science topic