A Pedestrian Sensitive Training Algorithm for False Positives Suppression in Two-Stage CNN Detection Methods

Pedestrian detection has been a hot spot in computer vision over the past decades due to the wide spectrum of promising applications, and the major challenge is false positives that occur during pedestrian detection. The emergence of various Convolutional Neural Network-based detection strategies substantially enhances pedestrian detection accuracy but still does not solve this problem well. This paper deeply analyzes the detection framework of the two-stage CNN detection methods and finds out false positives in detection results are due to its training strategy misclassifying some false proposals, thus weakening the classification capability of the following subnetwork and hardly suppressing false ones. To solve this problem, this paper proposes a pedestrian-sensitive training algorithm to help two-stage CNN detection methods effectively learn to distinguish the pedestrian and non-pedestrian samples and suppress the false positives in the final detection results. The core of the proposed algorithm is to redesign the training proposal generating scheme for the two-stage CNN detection methods, which can avoid a certain number of false ones that mislead its training process. With the help of the proposed algorithm, the detection accuracy of the MetroNext, a smaller and more accurate metro passenger detector, is further improved, which further decreases false ones in its metro passenger detection results. Based on various challenging benchmark datasets, experiment results have demonstrated that the feasibility of the proposed algorithm is effective in improving pedestrian detection accuracy by removing false positives. Compared with the existing state-of-the-art detection networks, PSTNet demonstrates better overall prediction performance in accuracy, total number of parameters, and inference time; thus, it can become a practical solution for hunting pedestrians on various hardware platforms, especially for mobile and edge devices.