Section 1 : Compute Lidar Point-Cloud from Range Image

Visualize range image channels (ID_S1_EX1)

The Visualized range image is

Visualize lidar point-cloud (ID_S1_EX2)

Sample of the point cloud image is

10 examples of vehicles with varying degrees of visibility in the point-cloud

There are many vehicle features that appear stable in most of the inspected examples. for example overall shape, rear bumper, bonnet. In some examples, tyre and rear lights are also good vehicle features.

Section 2 : Create Birds-Eye View from Lidar PCL

Convert sensor coordinates to BEV-map coordinates (ID_S2_EX1)

Compute intensity layer of the BEV map (ID_S2_EX2)

Compute height layer of the BEV map (ID_S2_EX3)

Section 3 : Model-based Object Detection in BEV Image

Add a second model from a GitHub repo (ID_S3_EX1)

Extract 3D bounding boxes from model response (ID_S3_EX2)

Section 4 : Performance Evaluation for Object Detection

Compute intersection-over-union between labels and detections (ID_S4_EX1)

ious: [0.9120538349785395, 0.8760647072678955]

center_devs: [[-0.0722, 0.0284, 0.8152483974641882], [0.0598, -0.0070, 0.8711946193956237]]

Compute false-negatives and false-positives (ID_S4_EX2)

det_performance: [[0.9120538349785395, 0.8760647072678955], [[-0.0722, 0.0284, 0.8152483974641882], [0.0598, -0.0070, 0.8711946193956237]], [3, 2, 1, 0]]

Compute precision and recall (ID_S4_EX3)

precision = 0.9691780821917808, recall = 0.9248366013071896

Result for configs_det.use_labels_as_objects

precision = 1.0, recall = 1.0