Fall Validation Performance

Fall System Requirements

Fall System Requirements

Test ID Description Peformance

(See Details Below)

INFRASTRUCTURE
Test 1 Detect single pedestrian centroid with Euclidean distance error < 0.3m Distance error < 0.09m
Test 2 Track single pedestrian within 20m of the infrastructure Success
Test 3 Predict single pedestrian trajectory 1.2 seconds into the future with an average error of 0.5m Average error = 0.46m.
Test 4 Time between first frame with a pedestrian to first published trajectory should be less than 1 second. Average time = 0.15 second.
VEHICLE
Test 5 Stop short of single pedestrian before he/she enters the field of view Success

 

Test 1: Detection Performance Test

Analysis testing: the unit test for detection module is conducted by assigning human agent to 16 of the pre-measured points. The errors are calculated as the Euclidean distance between the ground truth and detected coordinate. All the testing data are shown below:

Point No. Point Coordinate (m) Detected Coordinate (m) Error (m)
1 2,1 2.00,1.00 <0.0001
2 4,1 4.00,1.02 0.0200
3 6,1 5.99,1.02 0.0224
4 10,1 10.02,0.99 0.0224
5 15,1 15,0.98 0.0200
6 21,1 20.92,1.00 0.0800
7 21,3.3 20.85,3.34 0.1552
8 9,3.8 8.98,3.57 0.2309
9 8,3.8 8.04,3.72 0.0894
10 3,8 3.08,7.93 0.1063
11 3,6 3.05,5.94 0.0781
12 3,4 3.06,3.95 0.0781
13 3,2 3.03,1.98 0.0361
14 0,1 0.03,1 0.0300
15 -7,4 -6.86,4.09 0.1664
16 4,8 4.05,7.90 0.1118

 

The mean of Euclidean distance error is 0.0779m which proves that the detection module achieves a decent result and meets the FVE requirement.

 

Test 2: Detection Performance Test

Analysis testing: the unit test for tracking module is conducted by assigning human agent to walk between coordinate (20, 0) to (20, 3) repeatedly. The detected points’ y-coordinates versus time frame are stored and visualized. The optimal shape of the graph should be sine-wave like and y-coordinate value should be within the range of (-0.3, 3.3). The tracking test graph is shown below:

Figure 1. Tracking Test

 

Test 3: Prediction Performance Test

For the fall semester, we proceeded with a second degree polynomial fit. We tested this algorithm on live pedestrian data and were satisfied with the results that we were getting. The graphs below show the performance for some specific pedestrian trajectories.

Figure 2. Graphs of trajectory prediction for trajectories of different radii of curvature (r).
(a) r = ∞, (b) r = 3m, (c) r = 2m, (d) r = 1m, (e) r = 0m (right angle)

 

Test 4: Cycle Time Performance Test

The system takes 0.15 seconds between the first frame with a pedestrian to first published trajectory to the vehicle. Our requirement is <1 seconds.

 

Test 5: Single Pedestrian Collision Prevention Performance Test