Research Article

Traffic Flow Parameters Collection under Variable Illumination Based on Data Fusion

Table 2

Results of detection performance evaluation on the testing sets. On average of four testing subsets, the mean values of [Precision, F-measure] obtained by different methods are as follows: MOG [76.94%, 14.93%], MultiLayer [78.40%, 71.29%], CenterNet [86.17%, 89.97%], proposed method [88.03%, 91.10%].

Daytime-testing 1
MethodPrecisionRecallF-measureNFP/imageNFN/imagemAP

MOG83.27%9.83%16.76%0.1347.2174.86%
MultiLayer85.66%78.29%77.42%6.5915.6279.93%
CenterNet96.32%92.75%94.36%0.952.7997.57%
Proposed97.37%93.26%93.21%0.832.4698.1%

Daytime-testing 2
MOG86.18%9.33%17.14%0.5550.7373.24%
MultiLayer85.37%69.86%74.36%7.4817.2387.4%
CenterNet94.79%91.65%97.6%0.582.4998.2%
Proposed96.3%94.34%96.83%0.732.5297.93%

Nighttime-testing 1
MOG73.49%7.28%11.45%1.9043.6958.42%
MultiLayer74.28%61.16%62.13%3.512.2454.03%
CenterNet77.21%87.48%83.93%2.230.8575.99%
Proposed79.10%95.44%86.82%1.890.7588.39%

Nighttime-testing 2
MOG64.81%6.72%14.36%2.1350.7352.56%
MultiLayer68.27%64.62%71.26%2.872.6251.78%
CenterNet76.36%89.14%83.97%1.811.1075.37
Proposed79.33%93.12%87.54%1.341.8286.27%