Section: New Results

Late Fusion of Multiple Convolutional Layers for Pedestrian Detection

Participants : Ujjwal Ujjwal, François Brémond, Aziz Dziri [VEDECOM] , Bertrand Leroy [VEDECOM] .

One of the prominent problems in pedestrian detection is handling scale and occlusion. These problems are quite well aligned with the recent interests in autonomous vehicles. Successful detection of far-scale pedestrians can assist the vehicle in making safety maneuvers well ahead in time, thereby promoting a safer traffic environment. The same is true for surveillance systems in high security environment like airports and ports.

We propose a system design for pedestrian detection by leveraging the power of multiple convolutional layers explicitly (see Figure 5). We quantify the effect of different convolutional layers on the detection of pedestrians of varying scales and occlusion level. We show that earlier convolutional layers are better at handling small-scale and partially occluded pedestrians. We take cue from these conclusions and propose a pedestrian detection system design based on Faster-RCNN which leverages multiple convolutional layers by late fusion. In our design, we introduce height-awareness in the loss function to make the network emphasize on pedestrian heights which are misclassified during the training process. The proposed system design achieves a log-average miss-rate of $9.25\%$ on the caltech-reasonable dataset. This is within $1.5\%$ of the current state-of-art approach, while being a more compact system. The work was published in the 15^th IEEE International Conference on Advanced Video and Signal-based Surveillance (AVSS)-2018 [51].

Figure 5. Block diagram of our proposed Multiple-RPN pedestrian detection system