- FP.1 Match 3D Objects: L266 of
camFusion_Student.cpp. - FP.2 Compute Lidar-based TTC: L225 of
camFusion_Student.cpp. - FP.3 Associate Keypoint Correspondences with Bounding Boxes: L139 of
camFusion_Student.cpp. - FP.4 Compute Camera-based TTC: L176 of
FP.4 Compute Camera-based TTC.
Examples where the Lidar-based TTC estimate is way off - observations and argumentation.
All data observations:
LIDAR: median(prev)=8.074000, median(curr)=8.010000, TTC=12.515600 << [0] >>
LIDAR: median(prev)=7.947000, median(curr)=7.891000, TTC=14.091013 << [1] >>
LIDAR: median(prev)=7.891000, median(curr)=7.844000, TTC=16.689386 << [2] >>
LIDAR: median(prev)=7.844000, median(curr)=7.795000, TTC=15.908233 << [3] >>
LIDAR: median(prev)=7.795000, median(curr)=7.734000, TTC=12.678716
LIDAR: median(prev)=7.734000, median(curr)=7.670000, TTC=11.984351
LIDAR: median(prev)=7.670000, median(curr)=7.612000, TTC=13.124118
LIDAR: median(prev)=7.612000, median(curr)=7.554000, TTC=13.024118
LIDAR: median(prev)=7.554000, median(curr)=7.487000, TTC=11.174641
LIDAR: median(prev)=7.487000, median(curr)=7.429000, TTC=12.808601
LIDAR: median(prev)=7.429000, median(curr)=7.347000, TTC=8.959780
LIDAR: median(prev)=7.347000, median(curr)=7.274000, TTC=9.964390
LIDAR: median(prev)=7.274000, median(curr)=7.199000, TTC=9.598630
LIDAR: median(prev)=7.199000, median(curr)=7.116000, TTC=8.573525
LIDAR: median(prev)=7.116000, median(curr)=7.042000, TTC=9.516170
LIDAR: median(prev)=7.042000, median(curr)=6.969000, TTC=9.546581
LIDAR: median(prev)=6.969000, median(curr)=6.887000, TTC=8.398803While most of the observations are rather close to each other (no immediate outliers), a few do not necessary make sense in comparison to the rest, such as [1], [2] and [3] as marked above.
Outliers:
- Given the overall distance, the TTC of
[1-3]data points should be smaller than the very first observation (marked as[0]).
Observations:
median()function takes care of point clouds data outliers, so the average distance seems to be calculated correctly when compared to manual distance computation (from bird eye view LIDAR visualization).- The possible explanation is likely to be related to the chosen velocity model which is an approximation to the relative velocity changes.
- Generally, LIDAR measurements present a reliable source for TTC estimation even when using the velocity model.
All data observations:
| Detector | Descriptor | TTC |
|---|---|---|
| LIDAR | - | [12.52, 12.61, 14.09, 16.69, 15.91, 12.68, 11.98, 13.12, 13.02, 11.17, 12.81, 8.96, 9.96, 9.60, 8.57, 9.52, 9.55, 8.40, ] |
| BRISK | BRISK | [13.70, 25.27, 16.41, 17.22, 21.82, 18.82, 15.68, 15.78, 15.46, 13.55, 11.06, 12.09, 12.29, 12.20, 12.79, 11.44, 9.35, 11.90, ] |
| BRISK | BRIEF | [14.49, 12.34, 14.86, 17.78, 16.46, 16.67, 14.64, 14.40, 17.12, 11.56, 11.59, 13.25, 12.30, 11.39, 11.13, 9.60, 8.88, 10.41, ] |
| BRISK | ORB | [14.96, 23.20, 15.74, 16.69, 38.48, 27.38, 15.47, 17.37, 15.68, 12.72, 10.94, 13.29, 12.36, 12.66, 11.17, 11.57, 8.72, 12.17, ] |
| BRISK | FREAK | [13.82, 21.06, 13.10, 14.87, 23.05, 17.12, 19.55, 15.66, 16.61, 12.96, 11.97, 12.06, 12.05, 12.59, 12.60, 9.97, 8.61, 9.92, ] |
| BRISK | AKAZE | - |
| BRISK | SIFT | [15.76, 17.89, 18.39, 13.32, 34.55, 20.08, 15.07, 16.61, 17.21, 13.85, 13.63, 13.11, 14.78, 11.27, 14.76, 10.56, 9.66, 11.06, ] |
| ORB | BRISK | [16.73, -inf, 12.77, 18.06, 4008144.94, 18.12, -inf, 14.64, -inf, -inf, 8.51, -inf, 16.08, 20.05, 16.84, 12.30, 13.05, 36.31, ] |
| ORB | BRIEF | [20.18, -inf, 22.58, 14.13, 36.23, 10.13, -inf, -inf, -inf, 34.34, 24.59, 21.32, -inf, 20.17, 13.74, 9.37, 15.42, 16.81, ] |
| ORB | ORB | [12.18, -inf, 18.53, 30.38, -inf, -inf, -inf, -inf, -inf, 247.70, 8.02, -inf, 28.53, 38.83, 26.19, 16.84, 15.75, 245.09, ] |
| ORB | FREAK | [9.32, 47.92, 11.11, 11.37, -inf, 19.89, -inf, 9.53, -inf, -inf, 10.70, 22.56, 8.95, 54.18, 9.32, 7.31, 8.86, 8.82, ] |
| ORB | AKAZE | - |
| ORB | SIFT | [12.84, 20.89, 12.36, 44.29, -inf, -inf, -inf, 11.98, -inf, -inf, 8.29, -inf, 9.32, 17.72, 12.74, 9.50, 12.58, 24.87, ] |
| AKAZE | BRISK | [13.31, 13.92, 14.72, 14.98, 14.39, 15.34, 16.61, 14.11, 15.20, 13.20, 13.24, 11.15, 10.19, 9.80, 10.47, 11.77, 9.29, 9.02, ] |
| AKAZE | BRIEF | [13.59, 15.11, 13.06, 15.22, 15.06, 13.44, 16.18, 14.55, 14.24, 12.06, 12.77, 10.86, 10.31, 10.58, 10.07, 9.93, 9.51, 9.26, ] |
| AKAZE | ORB | [12.26, 13.50, 13.61, 14.18, 14.03, 13.77, 15.61, 14.11, 12.87, 12.38, 11.94, 10.81, 10.54, 12.76, 11.82, 10.61, 9.29, 8.43, ] |
| AKAZE | FREAK | [12.83, 13.81, 14.55, 15.53, 16.05, 16.29, 16.34, 13.45, 13.66, 12.08, 12.28, 11.23, 11.53, 10.07, 9.83, 9.61, 9.35, 9.07, ] |
| AKAZE | AKAZE | [12.45, 15.04, 13.06, 14.81, 14.70, 17.12, 15.74, 13.79, 14.61, 12.00, 12.27, 11.60, 11.08, 11.35, 10.00, 10.04, 9.42, 9.26, ] |
| AKAZE | SIFT | [12.66, 14.61, 13.66, 15.02, 15.16, 17.02, 15.41, 13.73, 14.52, 12.23, 12.24, 11.70, 10.70, 10.91, 10.93, 10.08, 9.18, 9.55, ] |
| SIFT | BRISK | [12.42, 13.11, 15.29, 17.56, 16.45, 11.86, 15.17, 14.70, 13.23, 11.66, 11.75, 10.62, 8.81, 9.13, 8.96, 8.36, 8.94, 9.02, ] |
| SIFT | BRIEF | [12.16, 14.82, 14.67, 17.77, 14.94, 13.74, 15.09, 14.89, 13.02, 11.52, 11.75, 11.11, 8.83, 9.48, 8.84, 8.13, 8.47, 9.91, ] |
| SIFT | ORB | - |
| SIFT | FREAK | [14.33, 13.33, 13.50, 19.29, 15.22, 11.74, 16.73, 15.24, 13.46, 12.17, 13.07, 10.45, 8.72, 9.65, 9.74, 8.34, 8.51, 10.51, ] |
| SIFT | AKAZE | - |
| SIFT | SIFT | [11.49, 12.87, 12.92, 17.11, 13.89, 12.19, 13.34, 14.18, 13.40, 10.66, 10.90, 10.62, 9.10, 9.22, 8.88, 8.44, 8.10, 7.99, ] |
| FAST | BRISK | [16.75, 11.94, 12.42, 14.97, 26.47, 13.50, 14.84, 14.20, 13.17, 15.42, 10.40, 12.65, 10.87, 11.39, 9.20, 9.85, 9.42, 9.54, ] |
| FAST | BRIEF | [12.22, 13.25, 12.34, 16.72, 16.08, 15.06, 16.99, 13.17, 14.49, 14.15, 11.69, 11.37, 10.20, 9.83, 10.70, 8.92, 9.27, 9.15, ] |
| FAST | ORB | [10.20, 12.45, 21.48, 13.79, 13.99, 13.37, 13.21, 13.15, 15.23, 15.02, 11.21, 11.98, 11.28, 10.04, 9.44, 9.27, 9.56, 10.33, ] |
| FAST | FREAK | [12.40, 12.37, 12.09, 15.49, 12.40, 12.27, 13.01, 12.35, 15.02, 13.30, 10.78, 12.45, 11.18, 11.06, 8.65, 9.57, 9.02, 9.16, ] |
| FAST | AKAZE | - |
| FAST | SIFT | [15.52, 11.38, 15.26, 38.28, 15.83, 13.41, 12.87, 13.99, 14.62, 16.55, 12.20, 11.87, 11.93, 10.91, 10.37, 9.93, 9.95, 10.25, ] |
Outliers:
- When the ratio is
1, it signifies there will never be collision (TTC =inf), or values can get pretty large for when vehicles move in similar velocities. - Potential reason: the ratio is a small difference and even small deviations create a large impact on the TTC estimation. Relying on ration requires highest possible precision when it comes to feature calculation. It is much safer to rely on LIDAR for exact this reason as the distance measurements are ingrained into pointcloud data nature.
Performance summary:
- SIFT-SIFT provided the best performance among all the detectors-descriptors pairs.
- Other best performing pairs are: FAST-FREAK and most of AKAZE-* pairs.
Selected best performing pairs vs. TTC of LIDAR:
