3D Multi Object Detection and Tracking using Matlab

Title: - A Novel Hybrid Deep Learning Approach based 3D Multi-Object Detection and Tracking from RGB-D and Fused LiDAR Point Clouds for Autonomous Driving
---------------------------------------------------------------------------------------------------------------------------------------------------------

Implementation Plan:
-----------------------

Step 1: Initially we load the input images from Solid-state LiDAR, Pseudo LiDAR, and RGB-D images.

Step 2: Next we perform the Solid-state LiDAR is the emerging technology in 3D object detection that overcomes the LiDAR in terms of low cost, high speed, and better accuracy. Next the preprocessing is classified into three stages they are,

2.1: Noise Removal: Noise Removal is performed using Adaptive Type-2 Fuzzy Filter (A-Fuzzy) algorithm with two stages. In the initial stage, the pixels are classified as noisy and good. The noisy pixels are taken in the second stage to remove the noise.
2.2: Contrast Enhancement: After removing the noise, contrast enhancement is performed to improve the quality of RGB-D images using the Moth Swarm Optimization (MSO) algorithm.
2.3: Points to Voxel Conversion: The enhanced LiDAR 2D cloud points are converted into 3D voxels for improving the perception view of the object to increase the detection accuracy of the 3D objects.

Step 3: Next we perform an L-GAN based Instance Segmentation process, In this process we rotate the images in terms of several degrees such as 10°, 90°, 180°, and 270°. After rotating the images, instance segmentation is performed for the fused images using Lightweight Generative Adversarial Network (L-GAN) algorithm.

Step 4: Nex we perform 3D Object Detection and Tracking process, In this process we perform feature extraction from the segmented region for classification of objects. Feature extraction and classification are performed using Hybrid Deep Learning algorithm which consists of YOLOv4 and VGG16 algorithms. After classification of objects, tracking is performed only for moving objects by considering RFID, unique ID, dimension, and orientation using Improved Unscented Kalman Filter (IUKF) algorithm.

Step 5: Finally, The performance of this research is evaluated in terms of following metrics,

5.1 Accuracy
5.2 Precision
5.3 Recall
5.4 F-measure
===============================================================================================================

Software Requirement:
-------------------------------

1. Tool: Matlab-R2020a
2. OS: Windows 10–(64-bit)
================================================================================================================

Note: -
---------

We perform the EXISTING process based on the REFERENCE 10 Title: - PointTrackNet: An End-to-End Network For 3-D Object Detection and Tracking From Point Clouds