December 2020

1. Parking Line Detection
2. Parking Slot Detection Software
3. Parking Slot Detection Github

Parking Line Detection

1. This was developed during one day and the algorithm is pretty stupid. Unfortunately the code is lost. However this is how it was done: 1. Perspective transfo.
2. Based on pillar information. Finally, the detected parking slots and free spaces are fused at a high-level to produce more reliable detection results. Once a target space is designated among the detection results, the parking slot tracking procedure is conducted. The target position is continuously tracked.
3. Parking occupancy detection. The key to active parking lot management is the reliable detection of available parking spaces. It is achieved by wireless sensor in every parking space, which can detect the presence of a car in the space and note the arrival/exit time of a vehicle.

tl;dr: Parking slot detection by detecting marking point with a CenterNet-like algorithm.

Overall impression

Sensors are mounted in the road and show the empty parking slot and allotted parking slot to the car parker. This sensor works like a Radar and Sonar. This sensor continuously transmitting and receiving waves. In parking-slot detection, we need to keep a track of the slot as we move from frame to frame. Cars will come and go their sizes will change and so will the bounding box. Mapping slots from frame to frame Intersection over Union (IOU) is a good metric in such problems.

For my future self: Dataset is super important. Your algorithm is only going to evolve to the level your dataset KPI requires it to.

The algorithm only focuses on detecting the marking point detection and did not mention too much about the post-processing needed to combine the marking points to parking slot. It is more general in that it can detect more than T/L-shaped marking points.

The paper is very poorly written, with tons of sloppy annotation and non-standard terminology.

Key ideas

• A coarse-to-fine marking point detection algorithm. Very much like CenterNet.
• The regression also predicts the “vertex paradigm”. Basically it predicts the pattern of the connectivity among the marking points.

Technical details

• Annotated a dataset (~15k images). This is slightly bigger than PS2.0 dataset with 12k images.
• The paper uses L2 loss to supervise the heatmaps and attributes. This is a bit strange as most studies uses focal loss for heatmap prediction and L1 for attribute prediction.

Parking Slot Detection Software

Notes

Parking Slot Detection Github

• Questions and notes on how to improve/revise the current work