Comparison of Mono Camera-based Static Obstacle Position Estimation Methods for Automotive Application
Summary (2 min read)
Introduction
- In the past decades automotive systems approach higher and higher levels of autonomy starting with cruise control (CC) until the first Level 2 autopilot systems (such as Tesla autopilot).
- Regarding stopping before a stationary target at 50km/h all systems performed well and stopped meanwhile at 80km/h six systems failed out of the ten.
- The research presented in this paper was funded by the Higher Education Institutional Excellence Program.
- Assuming known constant velocity the range of the obstacle can be estimated and with known range the size and side distance can also be estimated as pointed out in subsection III-A. Section III describes the selected four methods and extends them where possible, section IV evaluates the results of the test campaign and finally section V concludes the paper.
II. TEST SCENARIOS AND EVALUATION
- So either the center lines of the vehicles are aligned (center case) or there is one lane width difference between them (on the right side of the own, side case).
- These parameters are considered in the simulations.
- Additionally, pitch or yaw angle sinusoidal disturbances were applied to test the camera angular alignment sensitivity of the methods.
- Difference of estimated and real obstacle size and side distance are also calculated and compared at this time.
III. OVERVIEW OF METHODS
- Methods referenced in the introduction are collected here and extended to use N data points to smooth results and consider time-varying velocity if possible.
- At first, all methods are presented for constant velocity then the extensions are described.
A. Time to collision from scale change (SC)
- Probably the simplest method is the one published in [6].
- Unfortunately, the case of time-varying velocity can not be included into this framework as it does not comply with TTC estimation.
C. Distance estimation based-on ground contact point (G)
- It is based-on the known height of the camera mounted on the own car H and the vertical coordinate of the ground contact point of the obstacle in the camera image ygk (see Fig. 2).
- The distance can be directly calculated from them: Ẑk = fVH ygk (9) After determining Ẑk Ŵ and X̂0 can be determined similarly as in (2).
- This method does not include any assumption about the speed so time-varying speed does not cause problem.
D. Method based-on TTC and CPA estimation (SAA)
- This method was developed by the author for aircraft sense and avoid application as published in [12] and [13].
- At least two points are required to do this but more points will give better results as the tests will show.
- ˆCPA = X0W can be simply calculated as the average of the xk/Sk ratios.
- Extension of this method to time-varying speed is possible but requires a completely different solution method.
- It is advisable to multiply the first equation with fH to make it better conditioned and to solve it first for Z0/W, 1/W considering multiple measured points.
IV. TEST RESULTS
- First, the applicability of the constant velocity formulas was tested simulating constant vehicle speed.
- The figures show the limit speeds of applicability of each method as bar plots.
- Considering the small yaw disturbance the SCD method becomes inapplicable and the applicability of SC and G become severely limited while the limits for the SAA method stay the same as for the pitch case.
- Second, the applicability of the variable velocity formulas was tested simulating variable vehicle speed.
- The second best is the SAA method applicable until 90km/h.
V. CONCLUSIONS
- This paper presents the comparison of four methods capable to estimate the distance, position and size of a steady obstacle from monocular images.
- Constant and variable (sinusoidal and ramp) speeds were all considered in the range of 20 to 130 km/h.
- All cases were also tested with pitch and/or yaw angle disturbance to test the sensitivity of the methods to attitude disturbances (or camera misalignment).
- The evaluation criteria was the hitting speed of the obstacle (if the emergency braking does not stop completely the own vehicle because of estimation errors) and the precision of side distance and size estimation.
- The highest approach speed with which all defined thresholds are satisfied becomes the maximum applicability speed of each method.
Did you find this useful? Give us your feedback
Citations
30 citations
Additional excerpts
...While the seat-integrated techniques (ECG, cECG and BCG) have been used extensively in the automotive environment [121]; camera-based techniques, although not studied at a high level of readiness to the same extent as seat-integrated techniques, also remained popular [122,123]....
[...]
8 citations
Cites methods from "Comparison of Mono Camera-based Sta..."
...The braking model for the own vehicle is the same as in [7] considering the hardware delay (0....
[...]
...Mono camera-based distance estimation is relatively easy for static obstacles as for example [3] and the authors’ previous work [7] show....
[...]
References
306 citations
"Comparison of Mono Camera-based Sta..." refers background in this paper
...Scale change of the obstacle image and system sampling time are considered in [6] and [9] to estimate time to collision (TTC)....
[...]
...The authors target to compare existing mono camera-based methods without radar assistance as this is an extensively researched field in automotive ([2],[3],[4],[5],[6],[7],[8],[9]) and a good opportunity to evaluate their mono camera-based obstacle avoidance algorithm ([10],[11]) used in aerospace until now....
[...]
263 citations
"Comparison of Mono Camera-based Sta..." refers background in this paper
...Point of contact of the vehicle with ground and known camera height are considered in [2], [12] from which the range can be directly obtained and again the size and side distance....
[...]
...The authors target to compare existing mono camera-based methods without radar assistance as this is an extensively researched field in automotive ([2],[3],[4],[5],[6],[7],[8],[9]) and a good opportunity to evaluate their mono camera-based obstacle avoidance algorithm ([10],[11]) used in aerospace until now....
[...]
239 citations
"Comparison of Mono Camera-based Sta..." refers methods in this paper
...A braking model is set up in [14] giving the jerk as aa = −20m/s(3) and the brake system delay as t2 = 0....
[...]
60 citations
"Comparison of Mono Camera-based Sta..." refers background in this paper
...Scale change and traveled distance are considered in [8] from which the range and again the size and side distance can be determined....
[...]
...The authors target to compare existing mono camera-based methods without radar assistance as this is an extensively researched field in automotive ([2],[3],[4],[5],[6],[7],[8],[9]) and a good opportunity to evaluate their mono camera-based obstacle avoidance algorithm ([10],[11]) used in aerospace until now....
[...]
53 citations
"Comparison of Mono Camera-based Sta..." refers methods in this paper
...The braking model in [15] was modified to have three parts: distance traveled during system delay...
[...]
...A more detailed braking model is used in [15] as: S = (t1 + t2 + 0....
[...]
Related Papers (5)
Frequently Asked Questions (14)
Q2. What have the authors stated for future works in "Comparison of mono camera-based static obstacle position estimation methods for automotive application" ?
The first contribution of the paper is to extend the methods to consider N data points and variable velocity, however only three methods could be extended for variable velocity. Future work should include closed form sensitivity analysis of the methods to underline these results, consideration of measurement errors regarding velocity and vehicle position and the extension of the methods for moving obstacle if possible.
Q3. What is the added longitudinal velocity disturbance?
The added longitudinal velocity disturbance is: ∆Vz = 1.34 sin ( 2π T + π 2 ) [m/s] with T = 3sec period time whilethe lateral is: ∆Vx = 0.4 sin ( 2π T + π 2 ) [m/s].
Q4. What was the evaluation criteria for the SAA method?
The evaluation criteria was the hitting speed of the obstacle (if the emergency braking does not stop completely the own vehicle because of estimation errors) and the precision of side distance and size estimation.
Q5. What are the angular disturbances used to test the camera angular alignment sensitivity?
pitch or yaw angle sinusoidal disturbances were applied to test the camera angular alignment sensitivity of the methods.
Q6. How is the applicability of the own method tested?
The applicability limits of the own method are 90km/h for 20km/h and 100km/h for 30km/h hitting speed limit with the constant velocity formulas, which decreases to 70km/h and 90km/h in the variable speed cases respectively.
Q7. What is the braking model in [16]?
The braking model in [16] was modified to have three parts: distance traveled during system delay S2 = V0t2, distance traveled until deceleration builds up S3 = V0t3 + aa t33 6 and distance traveled during constant deceleration S4 = V3t4 +ax t24 2 .
Q8. What is the way to estimate the TTC?
the case of time-varying velocity can not be included into this framework as it does not comply with TTC estimation.
Q9. What is the main conclusion from the test campaign?
The overall conclusion form the test campaign is that the own SAA method is the most reliable considering also the attitude disturbances which are inevitable with vehicles and even a good attitude compensation system can left 1◦ error in the parameters.
Q10. What is the simplest formula to calculate the side distance?
the side distance can be calculated as X0 = CPA ·WFirst, the applicability of the constant velocity formulas was tested simulating constant vehicle speed.
Q11. What should be the future work of the SAA method?
Future work should include closed form sensitivity analysis of the methods to underline these results, consideration of measurement errors regarding velocity and vehicle position and the extension of the methods for moving obstacle if possible.
Q12. What is the typical lane width for a car?
The considered vehicle widths in lane design are 1.75m for car and 2.55m for truck as shown in [14] and a typical lane width can be 3m.
Q13. What is the basic formula for calculating the distances between the two points?
The basic formulas relate forward (Z) and side distances (X) and the object size (W) to the measurable image parameters (x,S):1Sk = Zk fHW , xk Sk = Xk W(11)Considering Vx = 0 side velocity (Xk = X0 = const), Vz < 0 forward velocity and the camera projection model in Fig. 1:1 Sk = − Vz fHW TTCk = Vz fHW︸ ︷︷ ︸ a tk − Vz fHW tC︸ ︷︷ ︸ b xk Sk = CPA(12)The first expression gives a possibility to fit line on the points 1/Sk, tk and so obtain the absolute time of collision as tC = −b/a and ˆTTCk = tC − tk.
Q14. What is the method to evaluate the distance of a vehicle?
Another test was the evaluation of variable speed formulas with ramp up (20 to 130km/h) or down (130 to 20km/h) vehicle speed profile reaching the end value at the obstacle.