A featureless approach for object detection and tracking in dynamic environments

Abstract

One of the challenging problems in mobile robotics is mapping a dynamic environment for navigating robots. In order to disambiguate multiple moving obstacles, state-of-art techniques often solve some form of dynamic SLAM (Simultaneous Localization and Mapping) problem. Unfortunately, their higher computational complexity press the need for simpler and more efficient approaches suitable for real-time embedded systems. In this paper, we present a ROS-based efficient algorithm for constructing dynamic maps, which exploits the spatial-temporal locality for detecting and tracking moving objects without relying on prior knowledge of their geometrical features. A two-prong contribution of this work is as follows: first, an efficient scheme for decoding sensory data into an estimated time-varying object boundary that ultimately decides its orientation and trajectory based on the iteratively updated robot Field of View (FoV); second, lower time-complexity of updating the dynamic environment through manipulating spatial-temporal locality available in the object motion profile. Unlike existing approaches, the snapshots of the environment remain constant in the number of moving objects. We validate the efficacy of our algorithm on both V-Rep simulations and real-life experiments with a wide array of dynamic environments. We show that the algorithm accurately detects and tracks objects with a high probability as long as sensor noise is low and the speed of moving objects remains within acceptable limits.

Design of an Improved Autonomous Tracked Vehicle Based on SLAM Algorithm

“…Based on SLAM algorithm improvement and optimization for local obstacle avoidance, the main research method is to achieve the positioning, mapping, and path planning functions of tracked robots in the ROS system platform [16].…”

A Cost-Effective Method for Detecting and Tracking Moving Objects using Overlapping Methods

“…In this step, several redundant and erroneous objects are eliminated from the results of the object detection process. A non-parametric background modeling approach [27,28] is chosen after choosing the moving item in the second phase to produce the initial detection results of the moving object with high recall but low precision. Target tracking is recovered in the third stage by utilizing an auto-encoder and deep learning techniques that are derivations of a hybrid method.…”

Key Indicators to Assess the Performance of LiDAR-Based Perception Algorithms: A Literature Review

“…Recent developments in image processing and machine learning techniques make it simpler to implement these tasks. Object detection and tracking [8], object classification [9], semantic segmentation/instance segmentation [10], and localisation [11] are eventually the most useful operations for the perception of vehicle surroundings. The perception and motion planning modules are the most difficult assignments.…”