Feature selection for Lidar-based gait recognition

Abstract

In this paper, we present a performance analysis of various descriptors suited to human gait analysis in Rotating Multi-Beam (RMB) Lidar measurement sequences. The gait descriptors for training and recognition are observed and extracted in realistic outdoor surveillance scenarios, where multiple pedestrians walk concurrently in the field of interest, their trajectories often intersect, while occlusions or background noise may affects the observation. For the Lidar scenes, we compared the modifications of five approaches proposed originally for optical cameras or Kinect measurements. Our results confirmed that efficient person re-identification can be achieved using a single Lidar sensor, even if it produces sparse point clouds.

Lidar-Based Gait Analysis and Activity Recognition in a 4D Surveillance System

“…As the last three columns of Table I confirm, the MLP and CNN outperformed each others on a case-by-case basis, and the committee has generally resulted in improved results over the two network components. As already shown in [39], in LGEI classification the MLP-CNN committee could also outperform the standard Vector Comparison approach proposed in [14].…”

“…Although pedestrian detection and tracking tasks have already been conducted on RMB Lidar measurements [10], [37], to our best knowledge our research [38], [39] has been the first attempt to involve such sensors in gait recognition. Due to the low spatial resolution of the sensor, and the presence of partially incomplete pedestrian shapes due to various occlusion effects, we decided to follow a model free approach, in contrast to model based methods [40], [41], [42] which fit structural body part models to the detected objects and extract various joint angles or body segment length parameters.…”

Evaluation of Cnn Architectures for Gait Recognition Based on Optical Flow Maps

“…A more recent work, [WHW15], uses a random set of binary silhouettes from a sequence to train a CNN that accumulates the calculated features to achieve a global representation of the dataset. complex work is presented in [GB15] where GEI are used to train an ensemble of CNN and a Multilayer Perceptron is employed as classifier. In [Wu17], given two GEI descriptors, they learn a metric to decide whether both descriptors belong to the same subject or not.…”

Gait Recognition with Compact Lidar Sensors

“…In (Gálai and Benedek, 2015) many state-of-the-art image based descriptors were tested for RMB LiDAR point cloud streams, proposed methods for both optical images (Kale et al, 2003) and point clouds were evaluated. (Tang et al, 2014) uses Kinect point clouds and calculates 2.5D gait features: Gaussian curvature, mean curvature and local point density which are combined into 3-channel feature image, and uses Cosine Transform and 2D PCA for dimension reduction, but this feature needs dense point clouds for curvature calculation, thus not applicable for RMB LiDAR clouds.…”

“…LGEI proved to be the most effective feature for LiDAR-based gait recognition in (Gálai and Benedek, 2015). The LGEI adopts the idea of the Gait Energy Image (Han and Bhanu, 2006), by averaging sideview silhouettes in a full gait cycle, with some small yet significant alternations.…”