Description

A new low illumination pedestrian sequence (LIPS) dataset is collected under night scenes in campus of Wuhan University. The dataset includes 90021 images of 100 pedestrians captured by two non-overlapping cameras. The length of each image sequence ranges from 193 to 833 frames,with an average number of 450 for each person at each camera. Many images are captured under very low illumination. Most of persons are occluded by other objects, e.g., other pedestrians, cars or trees. The illumination of some frames in persons video is severe low or huge changed by the cars light. All image sequences are normalized to 64 * 128 pixels after segmenting the person from each video.

Due to the limitation of page space and network bandwidth, we select 9 frames (1 frame every 10 frames) of each person in two cameras respectively, and show the small frames(32* 64 pixels) which are down-sampled with 1/2. As shown the following figure, the same person from two cameras are in each row. Left is from camera A. Right is from camera B. The colors and visual appearance of some persons are similar, e.g., person No. 38 and 57.

The illustration of self-developed software is shown following figure, which is running in .netframework4.0. Our software can not only extract persons from the low-illumination person frame sequences in each video, but also work under other complex scenarios, e.g., low resolution and foreground clutters, etc.

Operation Instructions:

This software can provide two camera settings, and sort the file by natural attribute and name the segmented image files;

1. Open the video frames ( The video must be transformed into frame sequences.), and use the mouse to draw a rectangle on the target pedestrians;

2. Adjust the rectangle with the keys [W, A, S, D];

3. Switch the next frame with the key [N], and [J] can return to the latest frame;

Examples

Downloads

Reference

1.Fei Ma, Xiaoke Zhu, Xinyu Zhang, Liang Yang, Mei Zuo, Xiao-Yuan Jing: Low illumination person re-identification. Multimedia Tools Appl. 78(1): 337-362 (2019)        

Literature