A Modified Pedestrian Retrieval Method Based on Faster R-CNN with Integration of Pedestrian Detection and Re-Identification
Enjia Chen, Xianghong Tang, Bowen Fu · 2018
In order to measure the similarity between two feature vectors more accurately, a linear combination of absolute difference and product of elements of two vectors is used in this paper to optimize distance function, along with a modified pedestrian retrieval framework based on Faster R-CNN. To get a more accurate pedestrian retrieval result, our proposed algorithm uses Region Proposal Network (RPN) to obtain candidate pedestrians, and the modified mixed similarity distance function to enhance the learning ability on similarity of features. The experiments on dataset CUHK-SYSU show that the proposed method achieves 80.9% in CMC top-1 and 78.8% in mAP, improving about 2.0%~18.0% in CMC top-1 and 3.0%~23.0% in mAP comparing to traditional methods.